Model Selection for Coding Tasks

This is a maintained, opinionated guide to choosing models for coding, review, and agentic work; prices and benchmarks below are as of 2026-11-08.

Model selection for coding is an economics decision before it is a capability decision: run the workhorse tier for the loop, buy the frontier by the minute, and give the reading to the cheap models. The part most engineers get backwards: for anyone paying per token, the harness you run moves your bill about as much as the model you pick, and sometimes more.

Task class decides the tier

#

Three task classes cover almost all coding work, and each maps to a price tier, not to a leaderboard rank.

• Completion and small edits: the model writes a few dozen lines from tight context; latency and price matter more than peak reasoning, so the cheap tier wins (gpt-6-luna and Claude Haiku 5.5 at $0.10/$0.50, Gemini 3.1 Flash-Lite, GLM-5.3-Flash).
• The agentic loop: explore, edit, test, repeat for minutes or hours; this is where most tokens die, and the mid tier (Claude Sonnet 5.5, gpt-6-sol, Gemini 3.1 Pro Preview) resolves most of it, while the Kimi, GLM, and DeepSeek challengers below price the same class far lower.
• Hard planning and review judgment: architecture choices, gnarly debugging, deciding what to let through; this is the only class where frontier spend reliably pays, and it is a small fraction of your turns.

Escalate by task class inside a session instead of picking one model for everything; Amp’s low/medium/high/ultra modes and Codex’s Sol/Terra/Luna defaults are this idea shipped as product (see the Amp and Codex notes).

The lineup as of 2026-11-04

#

The workhorse tier has converged to roughly $2 in and $10-12 out per million tokens at every major provider, which means switching costs are now measured in harness integration, not price. GPT-6 Sol joined the club on 2026-09-22 at exactly $2/$10, the first OpenAI model to price at the converged rate, while gpt-5.6-sol stays at $4/$20 behind its promotion. On 2026-09-29 OpenAI refreshed the tier again with gpt-6.1-sol at the same $2/$10 and the cache hit halved to $0.10, and on 2026-10-26 it filled the gap between the cheap tier and the workhorse rate with gpt-6.1-terra at $1/$5, the first GPT-6-family model priced between them. Claude added Opus 5.5 the same day at $4/$20 with a 5 percent cache-hit rate, the first Claude tier priced between Sonnet and Opus, and Claude added Sonnet 5.5 on 2026-09-28 at the unchanged $2/$10 workhorse rate, recorded at launch with the standard one-tenth cache hit and now billed at one twentieth, the same 0.05x multiplier as Opus 5.5. Google’s Gemini 4 Argon launched into the same tier on 2026-10-13 at $2/$10 with a 5 percent cache hit, a launch rate the pricing page schedules to double to $4/$20 on December 1. Moonshot put Kimi K3.1 Pro into the same tier on 2026-10-23 at $2/$10 with a $0.20 cache read, the first challenger flagship to price at the converged rate, while K3 holds $3/$15 and the K2 series that kimi-k2.7-code anchors is closed to new users from 2026-10-31 and slated to retire by the end of December 2026. gpt-5.6-terra and gpt-5.6-luna keep flipping on and off OpenAI’s pricing page, off on 2026-09-27, back on 2026-09-29, off again earlier on 2026-10-04, back on at launch prices by my second look the same day, listed at launch prices on 2026-10-07, off the page again when I re-checked on 2026-10-10, back on at launch prices by my 2026-10-11 re-check, still listed at my 2026-10-17 re-check, off the page again at my 2026-10-18 re-check, and back on at launch prices at my 2026-10-19 re-check, still listed there at my 2026-10-21, 2026-10-22, 2026-10-23, 2026-10-25, 2026-10-28, 2026-10-31, 2026-11-01, 2026-11-03, 2026-11-05, 2026-11-07, and 2026-11-08 re-checks, OpenRouter serving both at those rates throughout, Sol’s promo end extended from 2026-11-21 to 2026-11-30 at my 2026-11-05 re-check, OpenAI’s Daybreak line still carries gpt-5.6-cyber at $12.50/$75, and Gemini’s 3.8 Flash Cyber sells at $1.50/$7.50, the standard rate the regular Flash family only moves to on 2027-01-01.

A ? marks a property the verified pricing pages do not state. The lineup keeps only models released within the last year; Claude Haiku 4.5 left it when its one-year window closed on 2026-10-15, making gpt-5.3-codex (2026-02-05) the oldest entry, and release dates and context windows follow the models.dev list reference, which also supplies the newest Gemini Flash siblings the table groups on shared intro pricing, and keeps deepseek-flash’s release date at the line’s 2026-09-10 launch, the date DeepSeek’s own announcement carries, even though models.dev re-stamped the id at the 2026-10-17 V4.1-Flash-1017 checkpoint.

Five details the table hides:

• Batch mode is 50% off at OpenAI, Anthropic, Google, and Qwen3.8-Max, and Moonshot bills batch at 60% of standard on K3 and K3.1 Pro (stated on its pricing page as of 2026-11-03), which makes overnight review sweeps at least 40% off by default at five providers.
• Cached reads run about one tenth of input price at OpenAI, Google’s Gemini 3 family, and Kimi K3 and K3.1 Pro, one twentieth on Google’s new Gemini 4 Argon ($0.10) and on Anthropic’s Opus 5.5 and Sonnet 5.5, one tenth on the rest of Anthropic’s lineup except the Fable 5.1 and Mythos 5.1 pair at one fortieth, closer to one fifth at GLM and on kimi-k2.7-code, Qwen’s model tables name the discount while its billing note prices cache hits at one tenth of input, and DeepSeek bills hits at about one fiftieth on Flash and one thirtieth on V4 Pro, so loop economics depend on both the cache-hit rate and the vendor’s cache discount.
• Claude models from 4.6 onward include the full 1M-token context at standard pricing, Haiku 5.5 excepted above a 100K prompt, the cheapest flat-1M input prices belong to the challengers (deepseek-flash at $0.30 peak, GLM-5.3 at $1.40, Qwen3.8-Max and Kimi K3.1 Pro at $2, Kimi K3 at $3), and prices double past the long-context thresholds at OpenAI and on Gemini 3.1 Pro.
• The Claude 4.7+ tokenizer produces about 30% more tokens for the same text, so cross-vendor price comparisons understate Anthropic’s effective cost by roughly that margin.
• OpenRouter’s -pro GPT ids are a serving tier, not a premium model: the current-generation ones (the gpt-5.6 and gpt-6 families’ -pro ids) run the same underlying model with reasoning.mode set to pro and bill at the base id’s per-token rate, so the pro pass multiplies the bill in reasoning tokens at the same rate, unlike the legacy pattern where pro meant a separately priced tier (gpt-5-pro at $15/$120, gpt-5.5-pro at $30/$180).

The challengers reset the price floor

#

Kimi, GLM, DeepSeek, and Qwen price the agentic loop at half the converged workhorse rate or less, which makes the $2/$10-12 “standard” a choice rather than a fact. GLM-5.3, 5.2, and 5.1 sit at $1.40 in and $4.40 out per million tokens, with GLM-5 at $1/$3.20, per Z.ai’s pricing page as of 2026-09-27. Kimi’s coding-specialized kimi-k2.7-code lists at $0.95/$4.00 with a 256K context and multimodal input, and Moonshot’s own pricing and its official OpenRouter endpoint agree on that price as of 2026-09-25; the lease is now dated, Moonshot closes the K2 series to new users on 2026-10-31 and plans to retire all K2 models by the end of December 2026. Third-party OpenRouter hosts serve the same model id from $0.60/$3.00, but that floor is hosts undercutting and re-pricing weekly, not a Moonshot list price. Its HighSpeed variant lists at $1.90/$8.00 for about 180-260 tokens per second of output. Moonshot’s flagship line moved on 2026-10-23: Kimi K3.1 Pro, a post-trained K3 with substantially larger post-training compute, launches at $2/$10 with cache reads at $0.20 and the same flat 1M context, the first Moonshot model priced at the converged workhorse rate (a HighSpeed variant lists at $4/$20, and the full open weights shipped on schedule, with Moonshot’s own page now presenting K3.1 Pro as its most capable open-weight model and models.dev and Hugging Face carrying the released weights). Kimi K3 takes the cheap flank of the pair: $3/$15 list, which Moonshot’s page and its official endpoint both state, always reasoning with a configurable effort, and a flat-price 1M context that only Claude otherwise offers, with host floors that split again: the input leg to Wafer’s re-priced fast row at $0.167/$16.00 (cache reads $0.166) and the output leg holding on Relace’s fp4 row at $0.1824/$9.3388 (cache reads $0.18), with Chutes’ mxfp4 row still at $1.85/$9.50 (cache reads $0.185) without either leg (as of 2026-11-08), for token-payers who accept cheap-host variance. Z.ai even keeps GLM-4.7-Flash free, which makes it the zero-dollar candidate for inline completion and routing experiments. GLM-5.3-Flash fills the gap between free and $1.40: its 50 percent launch promo expired at 24:00 on 2026-09-09 Singapore time exactly as scheduled, and Z.ai’s page billed the $0.15/$0.50 list price with cached reads at $0.03 when I re-checked on 2026-09-29. On 2026-09-18 Z.ai wedged a new rung between Flash and GLM-5.3: GLM-5.3-FlashX, a high-speed Flash serving option for coding and agent workflows at $0.37/$1.25 with cached reads at $0.075. DeepSeek V4 undercuts them all: v4-pro lists at $1.32 in and $3.96 out per million at peak, with every hour outside 01:00-04:00 and 06:00-10:00 UTC on weekdays billed at half, so off-peak runs pay $0.66/$1.98. The V4 Pro retirement announced on 2026-09-10 is off: DeepSeek’s pricing page said on 2026-09-12 that V4 Pro service continues past 2026-09-14 with billing unchanged, so this row is permanent again. On 2026-09-10 DeepSeek replaced v4-flash outright: deepseek-flash (V4.1-Flash) lists at $0.30/$1.20 peak and $0.15/$0.60 off-peak, adds native vision input, and both models keep a flat 1M context with 384K max output and thinking mode on by default. DeepSeek’s cache hits bill at $0.044 (pro) and $0.006 peak or $0.003 off-peak (flash) per million, the Flash hit about one fiftieth of a miss, the deepest cache discount of any vendor in this guide. It also exposes an Anthropic-format endpoint, so harnesses that speak the Claude API can point at DeepSeek directly. One new bill to watch: DeepSeek’s Search and Fetch server-side tool calls, which DeepSeek Harness, Claude Code, and Codex sessions already hit, passed their stated 2026-11-01 start, and the pricing page now carries a standing Search and Fetch Pricing section billing Search and Fetch at $0.75 per 1,000 successful requests each, the free-through-October wording gone from the page as of 2026-11-05. Qwen3.8-Max rounds out the field at $2 in and $6 out per million tokens with flat 1M pricing, at or below the converged workhorse rate at flagship quality, and Qwen3.8-Flash sits in the cheap tier at $0.15/$0.47. Alibaba’s Model Studio is not a native OpenCode provider the way Moonshot, Z.AI, and DeepSeek are, but it speaks the OpenAI-compatible format, so Qwen rides OpenRouter or any compatible endpoint. Harness fit is solved: OpenCode lists Moonshot AI, Z.AI, and DeepSeek as native providers, so the lean BYOK harness plus a challenger model is one /connect away. My disagreeable claim: for a token-payer, the rational default loop in late October 2026 is deepseek-flash, kimi-k3.1-pro, or GLM-5.3 through OpenCode, and the big-three workhorses are what you escalate to, not what you default to. Whether the challengers hold quality on your codebase is exactly what SWE-bench’s bash-only view and a two-week cost-per-merged-PR measurement are for; the economics alone no longer justify defaulting to the big three. The era of Gemini Flash’s $0.75 intro rate leading a price category is over: gpt-6-luna and Claude Haiku 5.5 bill $0.10/$0.50, GLM-5.3-Flash matches their output rate at $0.15/$0.50, and deepseek-flash beats all of them on input at $0.30 peak, $0.15 off-peak, with no expiry announced.

Benchmarks decay faster than prices

#

Benchmark points do not track price, and the last independent leaderboard that published cost per run makes the point brutally. The aider polyglot leaderboard froze in 2025 along with the tool’s maintenance (see the aider note), but its dated runs still teach the right lesson. In October 2025, DeepSeek-V3.2-Exp (Reasoner) scored 74.2% for $1.30 of API spend across the 225-exercise suite. In June 2025, o3-pro (high) scored 84.9% for $146.32: ten points more for 113 times the money.

For current rankings, SWE-bench Verified is the reference, and its default view now runs every model in the same minimal mini-SWE-agent bash environment precisely so harnesses stop inflating scores. The frontier of evaluation moved on again: CodeClash (November 2025) scores goal-oriented development and ProgramBench (May 2026) scores building artifacts from scratch. My rule: use benchmarks for relative history and cost-per-point for decisions, and never quote a vendor’s self-reported number, including the SWE-Rebench claim on Junie’s own site (see the Junie note).

The harness multiplies the price of everything

#

The measured spread between harnesses is larger than the frontier premium at OpenAI and most of it at Anthropic. The July 2026 proxy study recorded in the Claude Code note measured about 33k input tokens sent before the user’s prompt on a minimal task, mid-session cache re-writes up to 54x, and a 4.2x multiplier on a two-subagent fan-out. The same study, recorded in the OpenCode note, measured about 7k baseline tokens for OpenCode with byte-stable cache prefixes, making it about 3.7x cheaper on a matched pass/fail benchmark. Compare: Fable 5.1 costs 5x Sonnet 5 per token, and gpt-5.6-cyber costs more than 6x gpt-6-sol. So choosing OpenCode over Claude Code moves a token-payer’s bill more than choosing Sonnet 5 over Opus 5.5, and roughly matches skipping Fable 5.1 for Sonnet 5. Models are substitutable across providers; a harness with a heavy baseline taxes every model you point through it, forever.

Subscription or BYOK: pick by where your spend lands

#

Subscriptions win below a spend threshold and handcuff you above it, so the unit that decides is dollars of monthly token spend, not features. The subscription side: Claude Code is included from Claude Pro ($20 monthly) through Max 20x ($200/month) per the Claude Code note, and Codex is included from the ChatGPT Free tier upward per the Codex note, though usage meters in shared five-hour windows with wide published ranges (10-100 local Sol messages on Plus). Amp’s Megawatt at $20/month makes a linked ChatGPT subscription’s usage effectively unlimited at low and medium modes per the Amp note, the cheapest delegation entry if you already pay for ChatGPT. The BYOK side: OpenCode, Junie, Crush, and aider bill at provider rates with zero markup (Junie’s cloud credits, 10 per 30 days on AI Pro, make BYOK effectively mandatory for real work per the Junie note). The crossover math: $100 of Sonnet 5 API spend is 50M raw input tokens at the table’s $2/M rate, and closer to 20M total tokens once you blend in output-priced generation at agentic ratios, which one heavy week can eat; if you reliably run past that, subscriptions cap your downside until their limits bite. One fence to remember: Anthropic’s March 2026 legal requests removed Claude subscription login from OpenCode (see the OpenCode note), so an Anthropic subscription does not transfer to third-party harnesses, while OpenCode’s provider docs now sign a ChatGPT subscription in with zero setup; subscribe to the vendor whose harness you will actually run.

A decision guide

#

Everything above collapses into one sentence per situation, and every price claim in it is dated in the table above.

• Daily loop default: Claude Sonnet 5.5, gpt-6-sol, Gemini 3.1 Pro Preview, Gemini 4 Argon at its launch $2/$10 rate, or Kimi K3.1 Pro at the same $2/$10; pick by harness fit, they price the same, and Kimi K3 at $3/$15 official (the OpenRouter floors sit on Wafer’s $0.167/$16.00 input row and Relace’s $0.1824/$9.3388 output row, with cheap-host caveats) is the cheaper-token challenger option in that class.
• High-volume loops where cost leads: deepseek-flash at $0.30 peak and $0.15 off-peak with no expiry announced, gpt-6-luna at $0.10/$0.50, GLM-5.3-Flash at $0.15/$0.50, or Gemini 3.8/3.7/3.6 Flash at the shared $0.75/$3.75 intro rate (calendar the 2026-12-31 end).
• Price-floor loops, BYOK: deepseek-flash, kimi-k2.7-code, or GLM-5.3 through OpenCode, all native providers; nothing credible stays cheaper.
• Whole-repository reads: DeepSeek from $0.30 in, a 1M-context Claude or Qwen3.8-Max at $2, Kimi K3.1 Pro at $2, or Kimi K3 at $3 flat (host floors from $0.167 in on Wafer and $9.34 out on Relace); all five keep 1M pricing flat.
• Frontier minutes only: gpt-6-astra, gpt-5.6-sol, Claude Opus 5.5, or Claude Opus 5 for planning and stuck debugging; Fable 5.1 when nothing else resolves.
• Inline completion and edits: gpt-6-luna, Claude Haiku 5.5, Gemini 3.1 Flash-Lite, or GLM-5.3-Flash.
• Review passes and doc reading: cheap tier in batch mode, or a 1M-context Claude for whole-repo reads at standard price.
• Offline or air-gapped: local models through aider, Crush, or Junie BYOK.
• Already paying ChatGPT: run Codex, and consider Amp for delegated work.
• Already paying Claude: run Claude Code; do not expect that subscription to drive any other harness.

What changes fast and how to re-verify

#

Everything volatile in this guide is dated, and four clocks are running. GLM-5.3-Flash’s half-price launch promo expired at 24:00 on 2026-09-09 Singapore time as scheduled, and Z.ai’s page billed the $0.15/$0.50 list price when I re-checked on 2026-09-29. DeepSeek cancelled the planned 2026-09-14 routing of v4-pro requests to V4.1-Flash and V4 Pro keeps billing at $1.32/$3.96 peak, Sol’s promotional pricing now runs through 2026-11-30 (the page extended the end date from 2026-11-21 at my 2026-11-05 re-check), Gemini 4 Argon’s launch rate becomes $4/$20 on December 1, the Gemini Flash intro rate ends 2026-12-31, Moonshot closes the K2 series to new users on 2026-10-31 and plans to retire all K2 models by the end of December 2026, and benchmark relevance decays on roughly a quarterly cycle (CodeClash, then ProgramBench within six months). The fourth clock is impersonal: OpenRouter’s third-party host floors on the Kimi models moved again by my 2026-11-08 re-check (the kimi-k3 aggregator listing holding at $0.192/$16.00 with cache reads at $0.182 while its endpoint board split again, the input leg to Wafer’s re-priced fast row at $0.167/$16.00 with cache reads at $0.166 and the output leg holding on Relace’s fp4 row at $0.1824/$9.3388 with cache reads at $0.18, and the roster narrowing to 29, while kimi-k2.7-code’s floor held at $0.60/$3.00 on CoreWeave’s int4 row with its cache read at $0.12), so treat those floors as spot prices, the Kimi Code note keeps the day-by-day table, and anchor purchasing decisions to the list prices. Sonnet 5 shows the other direction: a scheduled September 2026 increase to $3/$15 was cancelled weeks before taking effect, so scheduled changes are announcements, not facts. On each refresh I re-fetch the seven provider pricing pages (OpenAI, Anthropic, Google, Moonshot, Z.ai, DeepSeek, Alibaba) plus OpenRouter’s models and endpoints APIs and the models.dev list, the SWE-bench leaderboards, and one current harness-overhead measurement, and I update the table above and the as-of date together. A fact that cannot survive that re-fetch gets deleted rather than hedged.

What to Do Next

#

• Set your harness default to the workhorse tier this week, and escalate to a frontier model only for planning steps and stuck loops.
• Measure cost per merged PR on your own repository for two weeks; that number outranks every benchmark cited here.
• If you pay per token, prefer a harness with byte-stable cache prefixes, because the 54x cache re-write failure mode costs more than any model choice.
• Route review sweeps through batch mode; half price at the four providers that offer it and 40% off at Moonshot is the cheapest quality improvement available.
• The 2026-09-10 GLM-5.3-Flash check resolved as predicted: the promo expired into the $0.15/$0.50 list price; calendar the 2026-11-21 and 2026-12-31 promo deadlines to re-run your economics then.
• If you point a loop at DeepSeek, schedule the long runs off-peak; half price by the clock is the deepest vendor discount in this guide and it stacks with caching.

Changes

#

• 2026-08-24 - Created as the queue essay on task-class tiers; added the Kimi, GLM, and DeepSeek V4 lineups the same day on owner request.
• 2026-08-26 - Clarified the $100 crossover math so it stopped reading against its own table and added DeepSeek to the challengers mention.
• 2026-09-04 - Folded Claude Fable 5.1 into the lineup table and updated the cache bullet for its reduced cache-hit rate.
• 2026-09-08 - Transposed the lineup table on owner request so models read as columns against attribute rows, values unchanged.
• 2026-09-09 - Recorded the kimi-k2.7-code price drop to $0.71/$3.50 on OpenRouter and replaced the HighSpeed 2x claim with dollar figures.
• 2026-09-10 - Added DeepSeek’s deepseek-flash pricing, retired V4-Flash, and added the 2026-09-14 v4-pro routing clock.
• 2026-09-12 - Recorded kimi-k3’s OpenRouter drop to $2.65/$13.28 and removed the cancelled Flash-routing clock.
• 2026-09-16 - Recorded OpenRouter’s kimi-k2.7-code cut to $0.71/$3.21 with cache hits at $0.18 while Moonshot’s own page still lists $0.95/$4.00.
• 2026-09-18 - Recorded OpenRouter’s kimi-k3 cut to $2.10/$10.95 with cache hits at $0.23 while Moonshot’s own page still lists $3/$15.
• 2026-09-20 - Recorded OpenRouter’s kimi-k3 cut to $1.70/$8.50 with cache hits at $0.17 while Moonshot’s own page still lists $3/$15, and added Z.ai’s GLM-5.3-FlashX, released 2026-09-18 at $0.37/$1.25, to the lineup table and the challengers section.
• 2026-09-24 - Replaced the verification preamble with a single as-of clause in the intro on owner request.
• 2026-09-25 - Added GPT-6 Sol ($2/$10) and GPT-6 Luna ($0.10/$0.50), released 2026-09-22 per models.dev and launched to a 1,762-point thread, to the lineup table; the workhorse paragraph, cheap-tier bullets, and the price-category sentence updated.
• 2026-09-25 - Re-grounded the Kimi rows on Moonshot’s official prices after OpenRouter’s endpoints API showed the previously recorded cuts were third-party host floors, not vendor list prices: kimi-k2.7-code back to $0.95/$4.00 (cache $0.19) and kimi-k3 to $3/$15 (cache $0.30), with the host floors named in prose and the re-verify list now including the endpoints API.
• 2026-09-25 - Corrected the GLM-5.3-FlashX cached-read cell from $0.075 to the $0.090 Z.AI serves, and fixed the context-behavior row where the gpt-6-astra cell (doubles past threshold) and the Haiku 4.5 cell (200K) had been transposed.
• 2026-09-26 - Linked the new Model Access Feature Matrix, the layer of gateways, coding plans, and flat subscriptions that packages the per-token prices this guide tracks.
• 2026-09-27 - Refreshed every provider price: added Claude Opus 5.5 (2026-09-22, $4/$20, 5% cache) as a column, recorded gpt-5.6-terra and gpt-5.6-luna leaving OpenAI’s pricing page plus the new gpt-5.6-cyber and Gemini 3.8 Flash Cyber, corrected GLM-5.3-FlashX’s cached read back to $0.075, moved the Kimi K3 host floor to $0.88/$4.90, fixed the context-behavior row (Haiku 4.5 to 200K, Sonnet 5 and the Gemini Flash group to 1M flat), and corrected the re-verify list to seven provider pages including Alibaba.
• 2026-09-29 - Added Claude Sonnet 5.5 (2026-09-28, $2/$10, standard one-tenth cache) as a column, corrected the gpt-5.6-terra and gpt-5.6-luna cells after both returned to OpenAI’s pricing page at their launch prices, and moved the Kimi K3 OpenRouter host floor to $0.40/$10.00 (Relace).
• 2026-10-02 - Added gpt-6.1-sol (released 2026-09-29 per models.dev, $2/$10 with the cache hit halved to $0.10) as a column, re-verified all seven provider pricing pages with zero other vendor moves, and moved the OpenRouter host floors to their current spot readings (kimi-k2.7-code $0.67/$3.35, K3 $0.38/$10.00 on Relace).
• 2026-10-02 - Moved the Kimi K3 OpenRouter host floor from $0.38/$10.00 (Relace) to $1.44/$14.00 (Wafer) and the aggregator listing to $2.70/$13.50 within the same day; Moonshot’s official $3/$15 and every other provider anchor re-verified unchanged.
• 2026-10-03 - Moved the Kimi K3 OpenRouter host floor again, from $1.44/$14.00 (Wafer) to $0.98/$13.00 (Relace) with InferenceNet at $0.99/$13.00; Moonshot’s official $3/$15, the aggregator listing ($2.70/$13.50), the k2.7-code floor ($0.67/$3.35), and all seven provider anchors re-verified unchanged.
• 2026-10-04 - Recorded gpt-5.6-terra and gpt-5.6-luna dropping off OpenAI’s pricing page a second time (OpenRouter still serves both at launch rates) and re-anchored the as-of date to today.
• 2026-10-04 - Moved the Kimi K3 OpenRouter host floor again, from $0.98/$13.00 (Relace) to $0.40/$13.00 (Relace) with InferenceNet at $0.72/$13.00 and the aggregator listing at $0.50/$13.00; Moonshot’s official $3/$15 (via its pricing page and endpoint), the k2.7-code floor ($0.67/$3.35), and every other provider anchor re-verified unchanged against all seven pricing pages, OpenRouter, and models.dev.
• 2026-10-06 - Moved the Kimi K3 OpenRouter host floor again, from $0.40/$13.00 (Relace) to $0.96/$13.00 (Relace, cache reads $0.45) with Wafer at $0.97/$14.00, InferenceNet at $0.99/$14.00, and the aggregator listing itself at $0.99/$14.00; gpt-5.6-terra and gpt-5.6-luna stayed listed on OpenAI’s pricing page at launch prices, Sonnet 5’s release date corrected to 2026-06-30 per models.dev, and Moonshot’s official $3/$15 (pricing page and endpoint), the k2.7-code floor ($0.67/$3.35), and every other provider anchor re-verified unchanged against all seven pricing pages, OpenRouter, and models.dev.
• 2026-10-06 - Moved the Kimi K3 OpenRouter host floor a sixth time in six days, from $0.96/$13.00 (Relace, cache reads $0.45) to $0.66/$13.00 (Relace, cache reads $0.30), with InferenceNet and Wafer at $0.83/$14.00 and the aggregator listing itself moving to $0.83/$14.00; Moonshot’s official $3/$15 endpoint (cache $0.30), the k2.7-code floor ($0.67/$3.35, Inceptron), gpt-5.6-terra and gpt-5.6-luna still listed on OpenAI’s pricing page, and every other provider anchor re-verified unchanged this run.
• 2026-10-07 - Moved the Kimi K3 OpenRouter host floor a seventh time in seven days, from $0.66/$13.00 (Relace, cache reads $0.30) to $0.61/$13.00 (Relace, cache reads $0.45), with InferenceNet at $0.62/$15.00 (cache reads $0.43) and the aggregator listing itself moving to $0.62/$15.00; corrected the Qwen cache-hit claim, Alibaba’s billing page prices explicit-cache hits at one tenth of input and writes at 1.25x, so the rate is published; Moonshot’s official $3/$15 endpoint, the k2.7-code floor ($0.67/$3.35, Inceptron), gpt-5.6-terra and gpt-5.6-luna still listed on OpenAI’s pricing page, and every other provider anchor re-verified unchanged against all seven pricing pages, OpenRouter, and models.dev.
• 2026-10-10 - Added Claude Haiku 5.5 (2026-10-07 per models.dev, $0.10/$0.50 with cache reads at $0.01 for prompts up to 100K, prompt-length repricing to $0.50/$2.50 past it, 1M context) as a column, with Haiku 4.5 kept until it exits the one-year window on 2026-10-15, and updated the cheap-tier, price-category, and inline-completion mentions; recorded gpt-5.6-terra and gpt-5.6-luna dropping off OpenAI’s pricing page a third time (OpenRouter still serves both at launch rates).
• 2026-10-10 - Moved the Kimi K3 OpenRouter host floor an eighth time in eight days, from $0.61/$13.00 (Relace, cache reads $0.45) to $0.37/$14.89 (Wafer, cache reads $0.30), with Morph at $0.374/$14.90 (cache reads $0.29), InferenceNet at $0.64/$13.50 (cache reads $0.28), Relace falling to $2.00/$14.00, and the aggregator listing itself moving to $0.64/$13.50; Moonshot’s official $3/$15 (pricing page and endpoint), the k2.7-code floor ($0.67/$3.35), and every other provider anchor re-verified unchanged against all seven pricing pages, OpenRouter, and models.dev.
• 2026-10-11 - Moved the Kimi K3 OpenRouter host floor a ninth time in nine days, from $0.37/$14.89 (Wafer, cache reads $0.30) to $0.45/$16.50 (Wafer, cache reads $0.30), with InferenceNet at $0.73/$10.00 (cache reads $0.55), Morph at $1.00/$14.90 (cache reads $0.29), Relace at $2.00/$14.00, and the aggregator listing itself moving to $0.73/$10.00; recorded gpt-5.6-terra and gpt-5.6-luna back on OpenAI’s pricing page at launch prices a day after their third drop (OpenRouter served both at the same rates throughout); Moonshot’s official $3/$15 (pricing page and endpoint), the k2.7-code floor ($0.67/$3.35), and every other provider anchor re-verified unchanged against all seven pricing pages, OpenRouter, and models.dev.
• 2026-10-13 - Moved the Kimi K3 OpenRouter host floor from $0.45/$16.50 (Wafer, cache reads $0.30) to $0.73/$13.50 (InferenceNet, cache reads $0.55), with Wafer’s cheap endpoint at $0.78/$24.00 (cache reads $0.30) and the aggregator listing at $0.73/$15.00 (cache reads $0.55); corrected the Sonnet 5.5 cache-hit cell from one tenth to one twentieth (5 percent, $0.10/MTok) per the pricing page’s current 0.05x multiplier footnote, and narrowed the subscription fence so the no-transfer rule names Anthropic, since OpenCode’s provider docs now sign ChatGPT subscriptions in directly; Moonshot’s official $3/$15 (pricing page and endpoint), the k2.7-code floor ($0.67/$3.35, Inceptron), and every other provider anchor re-verified unchanged against all seven pricing pages, OpenRouter, and models.dev.
• 2026-10-13 - Added Gemini 4 Argon (released 2026-10-13 per models.dev, $2 / $10 with cache reads at $0.10 and flat 1M context) as a column, with Google’s pricing page scheduling the launch rate to become $4 / $20 on December 1; the workhorse-convergence paragraph, the cache bullet, and the daily-loop default updated, and the Google pricing-page and models.dev references re-dated to this fetch.
• 2026-10-13 - Moved the Kimi K3 OpenRouter host floor’s output leg from $13.50 to $10.00 on the same InferenceNet endpoint (cache reads $0.55), with the aggregator listing matching at $0.73 / $10.00 (cache reads $0.55), Wafer’s cheap endpoint at $0.72 / $24.00 (cache reads $0.30), and Morph at $0.75 / $14.90 (cache reads $0.29); Moonshot’s official $3 / $15 (pricing page and endpoint), the k2.7-code floor ($0.67 / $3.35), gpt-5.6-terra and gpt-5.6-luna still listed on OpenAI’s pricing page at launch prices, and every other provider anchor re-verified unchanged against all seven pricing pages, OpenRouter, and models.dev.
• 2026-10-16 - Dropped the Claude Haiku 4.5 column at its one-year window’s close (released 2025-10-15, kept per the 2026-10-10 note until it exited, gpt-5.3-codex of 2026-02-05 is now the oldest entry) and re-anchored the as-of date to a full re-verification of all sixteen cited sources; gpt-5.6-terra and gpt-5.6-luna still listed on OpenAI’s pricing page, Sonnet 5.5’s one-twentieth cache hit confirmed against the pricing page’s 5 percent footnote, and models.dev shows no new releases from the seven covered providers since 2026-10-13.
• 2026-10-16 - Moved the Kimi K3 OpenRouter readings to their 2026-10-16 spot prices: the aggregator listing at $0.304/$11.936 (cache reads $0.21, the listing’s fifteenth recorded move), the endpoint floor by input on Wafer at $0.29/$28.00, Relace at $0.35/$14.00, and Morph at $1.00/$14.90; Moonshot’s official $3/$15 (pricing page and endpoint), the k2.7-code floor ($0.67/$3.35), and every other provider anchor re-verified unchanged.
• 2026-10-17 - Moved the Kimi K3 OpenRouter readings to their 2026-10-17 spot prices: the aggregator listing held at $0.304/$11.936 (cache reads $0.21) while the endpoint floors churned, the input floor moving to Relace at $0.27/$14.2307 and the output floor to Sail Research’s fp4 endpoint at $0.504/$8.10, with Wafer at $0.28/$24.00 and $0.50/$14.90 and Morph at $0.30/$14.90 across a 26-endpoint roster; Moonshot’s official $3/$15 (pricing page and endpoint), the k2.7-code floor ($0.67/$3.35, Inceptron), gpt-5.6-terra and gpt-5.6-luna still listed on OpenAI’s pricing page, and every other provider anchor re-verified unchanged against all seven pricing pages, OpenRouter, and models.dev.
• 2026-10-18 - Re-anchored on a live re-check of all seven provider pricing pages, OpenRouter, and models.dev: gpt-5.6-terra and gpt-5.6-luna dropped off OpenAI’s pricing page a fourth time (OpenRouter still serves both at launch rates), the Kimi K3 endpoint board churned under a held aggregator listing ($0.304/$11.936, cache reads $0.21), the Relace floor row’s output leg moving from $14.2307 to $14.00, Wafer’s rows to $0.28/$16.00 and $0.36/$11.90, and Morph’s input leg to $0.365, while Moonshot’s official $3/$15 (pricing page and endpoint), the k2.7-code floor ($0.67/$3.35, Inceptron), and every other provider anchor re-verified unchanged, and models.dev re-stamped deepseek-flash at the 2026-10-17 V4.1-Flash-1017 checkpoint with the price unchanged and the table keeping the line’s 2026-09-10 launch date.
• 2026-10-19 - One closing reading for the daily-churn prices: gpt-5.6-terra and gpt-5.6-luna returned to OpenAI’s pricing page at launch prices a day after their fourth drop (OpenRouter serving both at the same rates throughout), and the Kimi K3 OpenRouter readings moved to their 2026-10-19 spot prices, the aggregator listing’s input leg dropping from $0.304 to $0.239 (sixteenth recorded move, output and cache legs held), the endpoint floor by input moving to Wafer at $0.238/$16.00 (cache reads $0.228), the output floor collapsing to Relace at $0.80/$3.00 after its wholesale reprice from $0.27/$14.00, InferenceNet’s fp4 row and the listing matching at $0.239/$11.936, Morph’s input leg down to $0.239, and Sail Research’s rows up to $1.20/$9.00 across the same 26-endpoint roster; Moonshot’s official $3/$15 (pricing page and endpoint), the k2.7-code floor ($0.67/$3.35, Inceptron), and every other provider anchor re-verified unchanged against all seven pricing pages, OpenRouter, and models.dev.
• 2026-10-21 - Delta-pass price check: gpt-5.6-terra and gpt-5.6-luna still listed on OpenAI’s pricing page at launch prices, and the Kimi K3 OpenRouter readings moved to their 2026-10-21 closing prices, the aggregator listing and the by-input floor holding ($0.239/$11.936, cache reads $0.21; Wafer $0.238/$16.00) while the output floor left Relace, whose one-day $0.80/$3.00 outlier re-priced to $0.30/$12.00, for Wafer’s second endpoint at $0.36/$11.90, with Sail Research’s output leg down to $5.50 and the roster widened to 28; the two daily-churn anchors held (the k3 listing at $0.239/$11.936, the kimi-k2.7-code floor at $0.67/$3.35, Inceptron), and every other provider anchor re-verified unchanged against all seven pricing pages, OpenRouter, and models.dev.
• 2026-10-23 - Delta-pass price check and a fifth hidden detail: gpt-5.6-terra and gpt-5.6-luna still listed on OpenAI’s pricing page at launch prices (OpenRouter serving both at the same rates), the Kimi K3 OpenRouter floors held at their 2026-10-21 readings (the listing at $0.239/$11.936 with cache reads at $0.21, the input floor at $0.238/$16.00 and the output floor at $0.36/$11.90 on Wafer’s endpoints, kimi-k2.7-code at $0.67/$3.35), and the endpoints roster widened from 28 to 29 with a Makora fp4 row at $0.65/$15.00 not previously recorded; added the OpenRouter -pro GPT-id cost trap to the hidden-details list, the -pro ids are serving tiers that bill at their base id’s per-token rate while spending far more reasoning tokens, verified id-for-id against OpenRouter’s models API from gpt-5.6-luna-pro to gpt-6.1-sol-pro.
• 2026-10-25 - Added Kimi K3.1 Pro (released 2026-10-23 per models.dev, $2/$10 with cache reads at $0.20 and flat 1M pricing, HighSpeed at $4/$20, full open weights due by November 4) as a column, the first Moonshot model at the converged workhorse rate, and stamped the K2-series sunset on the kimi-k2.7-code row, closed to new users 2026-10-31 with all K2 models slated to retire by end of December 2026; moved the Kimi K3 OpenRouter readings to their 2026-10-25 spot prices, the aggregator listing’s output leg down to $10.80 (listing $0.239/$10.80), the input floor to Relace at $0.19/$8.94, the output floor to Sail Research at $1.20/$5.50, roster steady at 29; gpt-5.6-terra and gpt-5.6-luna still listed on OpenAI’s pricing page, and the default-loop claim now names kimi-k3.1-pro beside deepseek-flash and GLM-5.3 with the K2 sunset noted; every other provider anchor re-verified unchanged against all seven pricing pages, OpenRouter, and models.dev.
• 2026-10-28 - Added gpt-6.1-terra (released 2026-10-26 per models.dev, $1/$5 with a $0.05 cache read) as a column, the first GPT-6-family model priced between the cheap tier and the converged workhorse rate, with astra’s newest-generation note moved to it; gpt-5.6-terra and gpt-5.6-luna still listed on OpenAI’s pricing page at launch prices, and the daily-churn Kimi readings moved (the kimi-k3 aggregator listing to $0.15/$20.00 with cache reads at $0.14, the input floor to Wafer at $0.169/$20.00, the output floor to Relace’s fp4 endpoint at $0.17/$8.099, the roster to 30, and kimi-k2.7-code’s host floor cut from $0.67/$3.35 to $0.56/$3.10 on Inceptron’s int4 row); every other provider anchor re-verified unchanged against all seven pricing pages, OpenRouter, and models.dev.
• 2026-10-29 - Delta-pass price check on the daily-churn anchors: the Kimi K3 OpenRouter readings moved (the aggregator listing to $0.192/$15.00 with cache reads at $0.182, both floors converging on one Relace kimi-k3-20260715 endpoint at $0.182/$9.34 after Wafer’s cheap input row left, the roster steady at 30), kimi-k2.7-code’s floor held at $0.56/$3.10 with its cache read at $0.15, and the five provider pricing pages whose last reading predated this week (Anthropic, Google, Z.ai, DeepSeek, Alibaba) re-fetched with every cell unchanged; the OpenAI pricing page (read 2026-10-28) and Moonshot’s pages carried no new moves.
• 2026-10-31 - Full seven-page sweep with the Moonshot batch term new on its pricing page: Moonshot now states batch inference at 60% of standard on K3 and K3.1 Pro, folded into the batch hidden detail and the batch recommendation; DeepSeek’s pricing page added per-request billing for its Search and Fetch server-side tool calls (free through October, $0.75 per 1,000 requests from 00:00 Beijing time on 2026-11-01 for DeepSeek Harness, Claude Code, and Codex sessions), recorded in the challengers section; the Kimi floors moved on OpenRouter (the K3 floor to $0.1495/$7.15 on Sail Research’s fp4 endpoint with the aggregator listing holding at $0.192/$15.00, kimi-k2.7-code’s floor to $0.60/$3.00 on CoreWeave’s new int4 row with Inceptron re-priced to $0.60/$3.50), gpt-6.1-terra confirmed listed on OpenAI’s own standard table, gpt-5.6-terra and gpt-5.6-luna still there at launch prices, and every other provider anchor re-verified unchanged against all seven pricing pages, OpenRouter, and models.dev.
• 2026-11-01 - Seven-page sweep plus the daily-churn Kimi readings: the kimi-k3 aggregator listing held at $0.192/$15.00 (cache reads $0.182) while its endpoint board moved, the input floor leaving the Sail Research fp4 row for a new Wafer fast row at $0.149/$22.00 (cache reads $0.148, Sail’s row still holding both legs at $0.1495/$7.15 and the cheapest single row), Relace’s former-floor row re-priced to $0.1824/$9.34, Makora’s fp4 row down to $0.50/$13.00, and the roster narrowing to 29, with kimi-k2.7-code’s floor held at $0.60/$3.00 on CoreWeave’s int4 row; the DeepSeek Search-and-Fetch billing reached its stated 2026-11-01 start with the pricing page still carrying the free-through-October wording, and every other provider anchor re-verified unchanged against all seven pricing pages, OpenRouter, and models.dev.
• 2026-11-03 - Seven-page sweep plus the daily-churn Kimi readings: the kimi-k3 aggregator listing held at $0.192/$15.00 (cache reads $0.182) while its endpoint board moved again, the Wafer fast row’s input floor re-pricing up from $0.149 to $0.152/$22.00 (cache reads $0.151), the output floor leaving Sail Research’s re-priced fp4 rows ($0.224/$10.72) for Wafer’s us row at $0.276/$9.60, Relace’s fp4 row now the cheapest single row holding both legs at $0.1529/$9.7471, Chutes’ $9.50 output row degraded, and the roster back at 30, with kimi-k2.7-code’s floor held at $0.60/$3.00 on CoreWeave’s int4 row; the DeepSeek pricing page still carried its free-of-charge Search-and-Fetch wording three days past the stated 2026-11-01 start, gpt-5.6-terra and gpt-5.6-luna still listed on OpenAI’s page at launch prices, and every other provider anchor re-verified unchanged against all seven pricing pages, OpenRouter, and models.dev.
• 2026-11-04 - Delta-pass price check on the daily-churn anchors: the Kimi K3 OpenRouter readings moved again (the aggregator listing re-priced from $0.192/$15.00 to $0.20/$16.00 with cache reads at $0.19, the input floor back to Relace at $0.19/$8.94 after Wafer’s fast row re-priced up to $0.233/$22.00 and Wafer’s us row re-priced its output to $11.50, Relace’s fp4 row again the cheapest single row, roster steady at 30) while kimi-k2.7-code’s floor held at $0.60/$3.00 on CoreWeave’s int4 row; the DeepSeek pricing page still carried its free-of-charge Search-and-Fetch wording four days past the stated 2026-11-01 start, gpt-5.6-terra and gpt-5.6-luna still listed on OpenAI’s page at launch prices with Sol’s promo still stated through November 21, and every other provider anchor re-verified unchanged (the five remaining pricing pages re-fetched 2026-11-03 with all cells unchanged, the Qwen billing page serving its Chinese locale this run with the batch and cache terms intact and the USD rates corroborated by OpenRouter and models.dev).
• 2026-11-05 - Delta-pass price check on the daily-churn anchors: the Kimi K3 OpenRouter readings split their floors across hosts (the aggregator listing’s input leg re-priced back down from $0.20 to $0.192 with the output held at $16.00 and cache reads at $0.182, Wafer’s fast row re-priced down to $0.153/$22.00 and reclaiming the input floor, Relace’s fp4 row re-priced up to $0.1539/$10.5539 and losing both legs, the output floor falling to Chutes’ mxfp4 row at $1.85/$9.50, no single row holding both legs, the roster widening to 31) while kimi-k2.7-code’s floor held at $0.60/$3.00 on CoreWeave’s int4 row; OpenAI’s page extended Sol’s promo end from 2026-11-21 to 2026-11-30 with gpt-5.6-terra and gpt-5.6-luna still listed at launch prices, DeepSeek’s page replaced its free-through-October wording with a standing Search and Fetch Pricing section at $0.75 per 1,000 requests five days past the stated start, and the K3.1 Pro open weights confirmed shipped (Moonshot’s page presents the model as open-weight with no pending-weights caveat, models.dev marks it Open, and the Hugging Face repo serves the release); every other provider anchor re-verified unchanged (Google, Anthropic, Z.ai, and Moonshot’s pages re-fetched with all cells unchanged, the Aliyun billing page again serving its Chinese locale with the batch and cache terms intact and the USD rates corroborated by OpenRouter and models.dev, which showed no new covered-provider releases after the 2026-10-26 terra rung).
• 2026-11-07 - Delta-pass price check on the daily-churn anchors: the Kimi K3 OpenRouter readings re-converged on Relace’s re-priced fp4 row at $0.1824/$9.3388 (cache reads $0.18), again the cheapest single row holding both legs, after Wafer’s fast row re-priced up to $0.192/$16.00 and Chutes’ mxfp4 row held at $1.85/$9.50 without either floor, with the aggregator listing holding at $0.192/$16.00 (cache reads $0.182) and the roster narrowing to 30, while kimi-k2.7-code’s floor held at $0.60/$3.00 on CoreWeave’s int4 row (its aggregator listing’s cache read moved to $0.16 at the held $0.60/$3.50 rates); gpt-5.6-terra and gpt-5.6-luna still listed on OpenAI’s page at launch prices with Sol’s promo still through November 30, and every other provider anchor re-verified unchanged against all seven pricing pages re-fetched this run plus models.dev, whose canonical Weights column read K3.1 Pro Closed against Moonshot’s open-weight card and the public Hugging Face release, recorded as a reference disagreement.
• 2026-11-08 - Delta-pass price check on the daily-churn anchors: the Kimi K3 OpenRouter readings split again (the aggregator listing holding at $0.192/$16.00 with cache reads at $0.182, the input floor to Wafer’s re-priced fast row at $0.167/$16.00 with cache reads at $0.166, the output leg holding on Relace’s fp4 row at $0.1824/$9.3388 with cache reads at $0.18, Chutes’ mxfp4 row still at $1.85/$9.50 without either leg, the roster narrowing from 30 to 29) while kimi-k2.7-code’s floor held at $0.60/$3.00 on CoreWeave’s int4 row (its aggregator listing holding at $0.60/$3.50 with the cache read at $0.16); gpt-5.6-terra and gpt-5.6-luna still listed on OpenAI’s page at launch prices with Sol’s promo still through November 30, and every other provider anchor re-verified unchanged against the pricing pages re-fetched this run plus models.dev, whose kimi-k3.1-pro entry still reads open_weights false against Moonshot’s open-weight card and the public Hugging Face release, the same reference disagreement.

See also

#

• Model Provider Feature Matrix - this guide’s prices regrouped into a provider-by-provider bundle comparison
• Model Access Feature Matrix - the gateways, coding plans, and flat subscriptions you can buy these models through
• Agentic Coding Tools Landscape - the four-layer map this guide’s harness claims sit inside
• Claude Code - the measured token overhead that anchors the harness-multiplier argument
• Codex - the Sol/Terra/Luna tiering as shipped defaults
• OpenCode - the lean BYOK counterexample and the subscription-transfer fence
• aider - the frozen leaderboard whose cost data this guide leans on

References

#

• https://platform.openai.com/docs/pricing - GPT-5.6 and GPT-6 family token prices including the 2026-09-29 gpt-6.1-sol ($2/$10, $0.10 cache read), batch discount, promo end date, long-context doubling; gpt-5.6-terra and gpt-5.6-luna still listed at launch prices at this fetch, with OpenRouter serving both at the same rates, and gpt-6.1-terra listed in the standard table ($1/$5, $0.05 cache read), re-confirmed at the 2026-10-31 fetch with Astra at $10/$50 and its $1.00 cache read; Sol’s promo end extended from 2026-11-21 to November 30, 2026 at the 2026-11-05 fetch (fetched 2026-10-22; re-fetched 2026-10-23, 2026-10-25, 2026-10-28, 2026-10-31, 2026-11-01, 2026-11-03, 2026-11-04, and again 2026-11-05, all cells unchanged each time except the Sol promo end date, gpt-5.6-terra and gpt-5.6-luna still listed at launch prices; re-fetched again 2026-11-08 with every cell unchanged, the promo still stated through November 30, and both models still listed at launch prices)
• https://platform.claude.com/docs/en/about-claude/pricing - Claude 5 lineup including the 2026-09-28 Sonnet 5.5 ($2/$10), the 2026-10-07 Haiku 5.5 ($0.10/$0.50 under a 100K prompt with $0.01 cache reads, prompt-length repricing past it), cache multipliers, 1M-context policy, tokenizer note, Sonnet 5 price decision (fetched 2026-10-22; re-fetched 2026-10-25, 2026-10-29, 2026-10-31, 2026-11-01, 2026-11-03, 2026-11-04, and again 2026-11-05, all cells unchanged; re-fetched again 2026-11-07, all cells unchanged; the docs.claude.com URL redirects here; re-fetched again 2026-11-08, all cells unchanged)
• https://cloud.google.com/vertex-ai/generative-ai/pricing - Gemini 3 family pricing, intro windows (the 2026-12-31 end and 2027-01-01 doubling both stated on the page), long-context and batch rates, plus Gemini 4 Argon’s launch table ($2 / $10 standard with $0.10 cache reads through November 30, then $4 / $20, the promo delivered as 50% credits back) (fetched 2026-10-22; re-fetched 2026-10-25, 2026-10-29, 2026-10-31, 2026-11-01, 2026-11-03, 2026-11-04, and again 2026-11-05, all cells unchanged; re-fetched again 2026-11-07, all cells unchanged with the Argon windows still stated; re-fetched again 2026-11-08, all cells unchanged with the Argon windows still stated and the page’s new Gemini 4 footnote scheduling retrievable-usage billing for completions over 32K tokens even on timed-out requests from November 13, 2026)
• https://www.swebench.com/verified.html - the Verified benchmark and the bash-only mini-SWE-agent comparison setup (fetched 2026-10-13, re-fetched 2026-10-25, unchanged; re-fetched 200 on 2026-11-08)
• https://www.swebench.com/ - CodeClash and ProgramBench launch dates, mini-SWE-agent 65% result (fetched 2026-10-13, re-fetched 2026-10-25, unchanged; re-fetched 200 on 2026-11-08)
• https://aider.chat/docs/leaderboards/ - the polyglot leaderboard with per-run costs and dates used for the cost-per-point argument (fetched 2026-10-13, re-fetched 2026-10-25, still frozen at its 2025 state; re-fetched 200 on 2026-11-08)
• https://platform.kimi.ai/docs/pricing/chat-k3 - Kimi K3 prices ($3/$15), cache-hit rate, flat 1M context, always-on reasoning with configurable effort (fetched 2026-10-22 through the page’s markdown render, whose tables the JS shell leaves empty; Moonshot’s official OpenRouter endpoint, re-read the same day, agrees; re-fetched 2026-10-25, 2026-10-28, and again 2026-10-31, the tables still render empty, the page carries the K2-series retirement notice and a Kimi K3.1 Pro entry, and its prose now states batch inference at 60% of standard on K3 and K3.1 Pro; re-fetched again 2026-11-01, 2026-11-03, 2026-11-04, and 2026-11-05 with the same shell, the same notice, and the batch term unchanged; fetched again 2026-11-07 with the same shell, the same notice, and the batch term unchanged; re-fetched 200 on 2026-11-08)
• https://platform.kimi.ai/docs/pricing/chat-k27-code - kimi-k2.7-code’s official prices ($0.95/$4.00, matching Moonshot’s own OpenRouter endpoint re-checked 2026-10-22, 2026-10-25, 2026-10-28, 2026-10-31, 2026-11-01, 2026-11-03, 2026-11-04, 2026-11-05, again 2026-11-07, and 200 on 2026-11-08), 256K context, HighSpeed variant prices and speeds (fetched 2026-10-22; re-fetched 2026-10-25 and again 2026-10-28, same shell, and the K2-series retirement notice on it names K3.1 Pro as the migration target)
• https://docs.z.ai/guides/overview/pricing - GLM-5.x family prices, cached input rates, free Flash tiers, the GLM-5.3-Flash list price billed after the 2026-09-09 promo, and GLM-5.3-FlashX’s 2026-09-18 price with its cache read (fetched 2026-10-22; re-fetched 2026-10-25, 2026-10-29, 2026-10-31, 2026-11-01, 2026-11-03, 2026-11-04, and again 2026-11-05, all cells unchanged; re-fetched again 2026-11-07, all cells unchanged; re-fetched 200 on 2026-11-08)
• https://platform.kimi.ai/docs/guide/kimi-k3-1-pro-quickstart.md - K3.1 Pro’s own page: a 3T-class open-weight post-train of K3 with a 1M-token context, flat pay-as-you-go pricing with no context-length tiering, and a HighSpeed variant (fetched 2026-10-25; its benchmark figures still render as placeholder brackets, so only the release, pricing, and capability facts above come from it, with the prices corroborated by models.dev and OpenRouter; the 2026-11-05 fetch presents the model as open-weight with the pending-weights caveat gone, matching the models.dev Open marker and the public Hugging Face release, which corroborate the weights’ on-schedule ship; fetched again 2026-11-07, still presenting the model as open-weight, with the Hugging Face release serving the weights; re-fetched 200 on 2026-11-08)
• https://opencode.ai/docs/providers/ - Moonshot AI, Z.AI, and DeepSeek as native OpenCode providers (fetched 2026-10-13, re-fetched 2026-10-25 with all three still listed, and re-fetched 200 on 2026-11-08; the page also documents ChatGPT Plus signing in with zero setup behind the subscription fence above)
• https://api-docs.deepseek.com/quick_start/pricing/ - DeepSeek lineup including V4.1-Flash and the notice that V4 Pro service continues past 2026-09-14 with billing unchanged, peak and off-peak rates, 1M context and 384K output, cache-hit prices, peak-hour definition, and the Search and Fetch per-request billing that starts 2026-11-01 (fetched 2026-10-22; re-fetched 2026-10-25, 2026-10-29, 2026-10-31, 2026-11-01, 2026-11-03, 2026-11-04, and again 2026-11-05, model cells unchanged each time; the 11-05 fetch replaced the free-through-October sentence with a standing Search and Fetch Pricing section billing Search and Fetch at $0.75 per 1,000 successful requests each, the free window’s wording gone five days past the stated 00:00 Beijing time start; the trailing slash matters: without it the URL serves the first-API-call page; re-fetched again 2026-11-07 with the standing section unchanged and the model cells unchanged; re-fetched again 2026-11-08 with the standing section and model cells unchanged)
• https://api-docs.deepseek.com/news/news260910 - the DeepSeek-V4.1-Flash release announcement of 2026-09-10: model id, V4-Flash retirement, the original 2026-09-14 V4 Pro routing plan since superseded by the pricing page’s continuation notice, and open weights
• https://help.aliyun.com/zh/model-studio/billing-for-model-studio - Qwen3.8-Max and Qwen3.8-Flash official per-token prices, batch half price on Max, context-cache discount, the 0-1M token tier, and the new-account free quota (fetched 2026-10-22; re-fetched 2026-10-25, 2026-10-29, 2026-10-31, 2026-11-01, 2026-11-03, 2026-11-04, and again 2026-11-05, price structure unchanged; the page’s context-cache note prices explicit-cache hits at one tenth of input and explicit-cache writes at 1.25x, the published rate the Qwen cache claims above now cite, and the 2026-11-04 and 2026-11-05 fetches served the page’s Chinese locale with CNY cells, batch half price and the cache discount intact, the USD rates corroborated those days by the OpenRouter listing and models.dev; the 2026-11-07 fetch served the Chinese locale again with the batch and cache terms intact, the USD rates corroborated by the OpenRouter listing and models.dev; re-fetched 200 on 2026-11-08)
• https://openrouter.ai/api/v1/models - aggregator listings for qwen3.8-max-0902 ($2/$6), qwen3.8-flash ($0.15/$0.47), the GPT-6 pair, gpt-6.1-sol ($2/$10), and kimi-k3.1-pro ($2/$10), re-checked 2026-10-29 through 2026-11-05 with every one of those rates unchanged; kimi-k2.7-code’s listing there is the third-party host floor (moved to $0.60/$3.50, cache reads $0.15, by the 2026-10-31 read, held at the 2026-11-01, 2026-11-03, 2026-11-04, and 2026-11-05 reads), not Moonshot’s list price, and kimi-k3’s listing re-priced from $0.192/$15.00 (cache reads $0.182, held 2026-10-29 through 2026-11-03) to $0.20/$16.00 (cache reads $0.19) on the 2026-11-04 read and back to $0.192/$16.00 (cache reads $0.182) on the 2026-11-05 read; gpt-5.6-terra and gpt-5.6-luna still served at their launch rates, and the -pro GPT ids on this same listing bill identically to their base ids (verified id-for-id 2026-10-23, gpt-6.1-sol-pro and gpt-5.6-luna-pro re-confirmed at the base rate 2026-10-29, the full current-generation -pro set including gpt-6.1-terra-pro and gpt-6-astra-pro re-confirmed at their base ids’ rates 2026-10-31, 2026-11-01, 2026-11-03, 2026-11-04, and again 2026-11-05; the 2026-11-07 re-check held the kimi-k3 listing at $0.192/$16.00 (cache reads $0.182), moved the kimi-k2.7-code listing’s cache read from $0.15 to $0.16 at its held $0.60/$3.50 prompt and output rates, and re-confirmed the -pro set at their base ids’ rates; the 2026-11-08 read held both Kimi listings at those rates and re-confirmed the -pro set)
• https://openrouter.ai/api/v1/models/moonshotai/kimi-k3/endpoints - per-provider endpoint prices separating Moonshot’s official $3/$15 (cache $0.30) from the floors: the 2026-11-05 read split the floors across hosts, Wafer’s fast row re-pricing down from $0.233/$22.00 to $0.153/$22.00 (cache reads $0.152) and taking the input floor, Relace’s re-priced fp4 row ($0.1539/$10.5539, cache reads $0.15) losing both legs it held on 11-04, Wafer’s us row re-pricing its output back down from $11.50 to $9.60, and the output floor falling to Chutes’ mxfp4 row at $1.85/$9.50 (cache reads $0.185), leaving no single row holding both legs, with the roster widening to 31 endpoints (fetched 2026-10-22; re-fetched 2026-10-23, 2026-10-25, 2026-10-28, 2026-10-29, 2026-10-31, 2026-11-01, 2026-11-03, 2026-11-04, and again 2026-11-05, the floor leaving Relace for Sail Research’s fp4 row on the 10-31 read before Wafer’s fast row split it, the board re-converging on Relace on the 11-04 read, and splitting again on the 11-05 read; the 2026-11-07 read re-converged the board on Relace’s re-priced fp4 row at $0.1824/$9.3388 (cache reads $0.18) holding both legs while Wafer’s fast row re-priced up to $0.192/$16.00 and the roster narrowed to 30; the 2026-11-08 read split the floors again, Wafer’s fast row re-pricing down to $0.167/$16.00 (cache reads $0.166) and taking the input leg while Relace’s fp4 row held the output leg at $0.1824/$9.3388, with the roster narrowing to 29)
• https://models.dev - the community model list used as the lineup reference: model ids, release dates, and context windows for every model in the table, including the GPT-6 pair’s and Opus 5.5’s 2026-09-22 release dates, Sonnet 5.5’s 2026-09-28, gpt-6.1-sol’s 2026-09-29, Haiku 5.5’s 2026-10-07, gemini-4-argon’s 2026-10-13, kimi-k3.1-pro’s 2026-10-23, and gpt-6.1-terra’s 2026-10-26 (fetched 2026-10-28, 2026-10-31, 2026-11-01, 2026-11-03, 2026-11-04, and again 2026-11-05 with no new covered-provider releases after terra; its deepseek-flash entry stays re-stamped at the 2026-10-17 V4.1-Flash-1017 checkpoint, and the 11-05 fetch marks kimi-k3.1-pro’s weights Open, corroborating the on-schedule ship Moonshot’s page no longer caveats; the 2026-11-07 fetch still shows no new covered-provider releases but reads kimi-k3.1-pro’s canonical Weights column Closed against Moonshot’s open-weight card and the public Hugging Face release, the disagreement recorded with the weights claim grounded on the primaries; the 2026-11-08 fetch shows no new covered-provider releases and the same open_weights-false reading standing)