On API list price, GPT-6.1 Sol is the cheapest of the three at $2 input and $10 output per million tokens; Gemini 4 Argon matches it until November 30, 2026 and then doubles to $4 / $20; Claude Opus 5.5 costs $4 / $20 today. Pick by where you already work: GPT-6.1 Sol for ChatGPT and Codex users, Opus 5.5 for Claude Code users whose tests justify twice the token price, and Gemini 4 Argon for Google AI Pro subscribers and Google Cloud teams.
All three shipped within three weeks: Claude Opus 5.5 on September 22, 2026, GPT-6.1 Sol on September 29, and Gemini 4 Argon became generally available on October 13. Strictly, OpenAI's top model is GPT-6 Astra and Anthropic's is Claude Fable 5.1 (both $10 / $50); we compare GPT-6.1 Sol because paid ChatGPT plans use it in ChatGPT Work and Codex and it lists at Argon's launch price. Every fact below was read on vendor pages on October 23, 2026. We do not publish our own benchmark ranking.
| GPT-6.1 Sol | Claude Opus 5.5 | Gemini 4 Argon | |
|---|---|---|---|
| Vendor | OpenAI | Anthropic | |
| Released | September 29, 2026 | September 22, 2026 | Generally available October 13, 2026 |
| Input / output per 1M tokens | $2 / $10 (prompts up to 272K) | $4 / $20 | $2 / $10 through November 30, 2026; $4 / $20 from December 1, 2026 |
| Cached input | $0.10 | $0.20 (cache read) | $0.10, then $0.20 (implicit caching only) |
| Batch | $1 / $5 | $2 / $10 | $1 / $5, then $2 / $10 |
| Faster tiers | Fast $4 / $20; Ultrafast $12 / $60 | Fast mode $8 / $40 | Not compared |
| Long prompts | $4 / $15 above 272K input tokens | See pricing page | See pricing page |
| Context window | 1,050,000 tokens | 1M tokens | 1,048,576 tokens shared by input and output |
| Max output | 128,000 tokens | 128K tokens | 65,536 by default; up to 1,048,576 with long decode |
| Reasoning control | Reasoning effort | Adaptive thinking | Thinking levels low, medium, high (default medium) |
| Knowledge cutoff | April 30, 2026 | June 2026 | Not compared |
Gemini 4 Argon is not on the Gemini API free tier. Above these sit GPT-6 Astra and Claude Fable 5.1, both at $10 / $50.
Which to use for what
| Job | Start with | Why |
|---|---|---|
| Coding inside an agent | The model your plan already includes: GPT-6.1 Sol in Codex, Opus 5.5 in Claude Code, Argon in GitHub Copilot | Plan limits, not token prices, decide how much you can use |
| Coding through the API at volume | GPT-6.1 Sol, or Gemini 4 Argon until November 30, 2026 | Half the per-token price of Opus 5.5 |
| Very long documents | Any of the three | All accept about a million tokens; Argon shares its window between input and output |
| Writing | Test all three on your own samples | No vendor-neutral test settles style; LMArena's text leaderboard listed Argon first and Opus 5.5 second on October 23, 2026 |
| Research in a consumer app | Gemini 4 Argon in Google AI Pro, or ChatGPT | Google AI Pro bundles Argon with Deep Research for $19.99 |
| Lowest cost over the next year | GPT-6.1 Sol | Argon's $2 / $10 is introductory and ends November 30, 2026; Sol's price is not marked as promotional |
Worked example: one month of API use, checked twice
Assume a team sends 20 million input tokens and receives 4 million output tokens in a month, all prompts under 272K tokens, at standard rates with no caching. Cost is input millions times the input rate plus output millions times the output rate.
| Model | Input | Output | Total | With Batch (50% off) |
|---|---|---|---|---|
| GPT-6.1 Sol | 20 × $2 = $40 | 4 × $10 = $40 | $80 | $40 |
| Claude Opus 5.5 | 20 × $4 = $80 | 4 × $20 = $80 | $160 | $80 |
| Gemini 4 Argon, through November 30, 2026 | 20 × $2 = $40 | 4 × $10 = $40 | $80 | $40 |
| Gemini 4 Argon, from December 1, 2026 | 20 × $4 = $80 | 4 × $20 = $80 | $160 | $80 |
| For scale: GPT-6 Astra or Claude Fable 5.1 | 20 × $10 = $200 | 4 × $50 = $200 | $400 | $200 |
Second check, by bundle: in this mix every 1 million output tokens comes with 5 million input tokens. On GPT-6.1 Sol one bundle costs $10 plus $10, so $20; on Opus 5.5 it costs $20 plus $20, so $40. The month is four bundles, so $80 and $160, matching the table. At list price, Opus 5.5 costs exactly twice GPT-6.1 Sol for the same tokens, and Argon moves from Sol's price to Opus 5.5's price on December 1, 2026.
Token counts are not identical across vendors: each uses its own tokenizer, and Anthropic notes that Claude models from 4.7 onward produce about 30% more tokens for the same text. Reasoning can also add output tokens. Measure on your own traffic before you set a budget.
Subscriptions: what about $20 a month?
| Plan | US price | What you get |
|---|---|---|
| ChatGPT Plus | $20/month | GPT-6 Sol in Chat; GPT-6.1 Sol in ChatGPT Work and Codex |
| Claude Pro | $20/month, or $17/month billed annually | Opus 5.5 included (Opus is not on Free); Claude Code |
| Google AI Pro | $19.99/month (€21.99 in Germany) | Gemini 4 Argon, Deep Research, 5 TB storage |
| GitHub Copilot | From $10/month | Gemini 4 Argon available since October 13, 2026 |
Each plan meters usage differently: ChatGPT and Claude combine 5-hour windows with weekly limits, and Google describes compute-based limits that refresh every 5 hours until a weekly limit.
What the leaderboards say, and what they do not
As listed on Artificial Analysis on October 23, 2026, Claude Opus 5.5 (max with fallback) led its Intelligence Index at 58, while Claude Fable 5.1, GPT-6 Astra, and Gemini 4 Argon (high) were listed at 53; GPT-6.1 Sol was not among the top entries we recorded. As listed on LMArena's text leaderboard the same day, Gemini 4 Argon (High) was first at 1517 (marked preliminary) and Claude Opus 5.5 (High) second at 1509. The two leaderboards disagree on the top spot and test specific settings, so use them to decide what to test, not what to buy.
Verdict
- Choose GPT-6.1 Sol if cost per token matters most, you already use ChatGPT or Codex, or you want a price that is not marked as introductory.
- Choose Claude Opus 5.5 if you work in Claude Code or the Claude apps and your own tests show it fixes failures the cheaper models miss. Budget twice Sol's token cost, or try Claude Sonnet 5.5 at $2 / $10 first.
- Choose Gemini 4 Argon if you already pay for Google AI Pro or build on Google Cloud, and budget on the $4 / $20 rate that starts December 1, 2026.
- Choose none of them for high-volume simple work: GPT-6 Luna and Claude Haiku 5.5 list at $0.10 / $0.50.
Compare every current model in one table.
Prices, context windows, and status for OpenAI, Anthropic, Google, and open-weight models.
FAQ
Which is cheapest: GPT-6.1 Sol, Claude Opus 5.5, or Gemini 4 Argon?
Until November 30, 2026, GPT-6.1 Sol and Gemini 4 Argon tie at $2 / $10 per million tokens and Opus 5.5 costs $4 / $20. From December 1, 2026, Argon moves to $4 / $20, leaving GPT-6.1 Sol the cheapest of the three.
Is Gemini 4 Argon free?
Not on the API: Google lists it as not available on the Gemini API free tier. In the Gemini app, Google lists Gemini 4 Argon as part of Google AI Pro ($19.99 a month in the US, €21.99 in Germany).
Is GPT-6.1 Sol the same as GPT-6 Sol?
No. GPT-6 Sol (released September 22, 2026) is now listed under "More models" on OpenAI's models page. Both list at $2 / $10, but GPT-6.1 Sol's cached input costs $0.10 versus $0.20, and its knowledge cutoff is April 30 versus April 20, 2026. ChatGPT Chat uses GPT-6 Sol on paid plans; ChatGPT Work and Codex use GPT-6.1 Sol.
Which is best for coding?
We do not run our own benchmark; our editorial picks by job are on /best-ai-model. In practice, the agent you use decides more than the model: Codex runs OpenAI models, Claude Code runs Claude models, and GitHub Copilot offers Gemini 4 Argon among its options. Run the same task in each and compare the diffs.
What happened to GPT-5.5?
OpenAI retired GPT-5.5 from ChatGPT, ChatGPT Work, and Codex on October 14, 2026. It remains in the API at $5 / $30 per million tokens, with no shutdown date on OpenAI's deprecations page as of October 23, 2026.