Noch nicht auf Deutsch – hier die englische Seite.

GPT-6 vs Claude Opus 5.5 (November 2026): which is better for coding, and what each costs

Quick answer

For coding, Claude Opus 5.5 and GPT-6 Astra trade wins on the published benchmarks, and Opus 5.5 costs less through the API: $4 per million input tokens and $20 per million output tokens, against GPT-6 Astra's $10 and $50 (checked November 16, 2026). Anthropic's Terminal-Bench 4.0 run puts Opus 5.5 ahead, 66.4% to 57.9%, and both Anthropic's and Google's FrontierSWE v2 tables put GPT-6 Astra ahead, 65.5% to 62.3%. Past 272K input tokens, Astra's rate rises to $20 / $75 for the whole request, while Opus 5.5 keeps one rate across its 1M-token context.

Publish my AI-made page free

Free plan: no credit card, 1 published page, 50 form entries a month.

On this page

You're picking the model for a coding agent or an app, and the 2 names on your list are GPT-6 and Claude Opus 5.5. GPT-6 is a family of OpenAI models, so this page puts Claude Opus 5.5 beside GPT-6 Astra, the one OpenAI's models page names first: "If you're not sure where to start, use GPT-6 Astra, our flagship model for complex reasoning and coding." GPT-6.1 Sol, OpenAI's cheaper model, joins where its price changes the answer. For Gemini, Grok, and the other labs, see the best AI models right now.

Every model fact, price, and benchmark below comes from OpenAI's, Anthropic's, and Google's own pages, read on November 16, 2026, between 14:41 and 14:50 Japan time. The sources list each page. Benchmark numbers appear as the lab that published them reports them, with that lab named, because the labs run the same tests in different setups and get different numbers. Where a lab's pages don't say something, the table says so. Prices are US dollars per million tokens, input / output.

GPT-6 vs Claude Opus 5.5 at a glance

Each cell links to the page it comes from, checked on November 16, 2026:

What the short cells leave out. Per token, Opus 5.5 costs 40% of Astra's price. Astra's higher rate applies to the whole request once its input passes 272K tokens, so a long prompt costs 2x on input and 1.5x on output. Both labs bill reasoning as output: OpenAI says reasoning tokens "are billed as output tokens," and Anthropic says the tokens Claude spends thinking "are billed as output tokens." A higher effort setting raises the output side of the bill on either model. On Astra, tool calls need the Responses API, and custom temperature or top_p values aren't supported. On Opus 5.5, adaptive thinking "is always on and can't be turned off."

GPT-6 vs Claude Opus 5.5 for coding

No lab has published its own run of both models on the same coding test, and OpenAI's GPT-6 Astra launch post returned an error when read for this guide. Here is every coding score that names both models, from Anthropic's and Google's tables:

BenchmarkReported byGPT-6 AstraClaude Opus 5.5
Terminal-Bench 4.0Anthropic, September 22 (Opus 5.5 at xhigh; Astra at high, as reported by OpenAI)57.9%66.4%
Terminal-Bench 4.0, run by Artificial AnalysisGoogle, Gemini 4 Carbon model card (results as of October 2026)59.1%59.6%
FrontierCode v1.1, main setAnthropic, September 2253.3%54.4%
FrontierCode 2.0Anthropic, November 942.8%45.5%
FrontierSWE v2Anthropic, November 9; Google's model card gives the same figures65.5%62.3%
SWE-sweepGoogle, Gemini 4 Carbon model card9.5%3.6%

Opus 5.5 leads 4 of these 6 rows and GPT-6 Astra leads 2. The widest gap is Anthropic's own Terminal-Bench 4.0 run, 8.5 points to Opus 5.5; on Google's run of the same test, the two are half a point apart. GPT-6 Astra leads FrontierSWE v2 by 3.2 points in both Anthropic's and Google's tables, and Google's SWE-sweep by 5.9.

Each lab also reports cost per task, in its own words:

  • Anthropic, September 22: at its default effort, Opus 5.5 scores 54.6% on FrontierCode v1.1, "beating GPT-6 Astra's top score (53.3%) for about a fifth of the cost per task." On Terminal-Bench 4.0, "it matches Astra for about 40% of the cost."
  • OpenAI, September 29: OpenAI's pages read for this guide don't compare Astra with Opus 5.5 on a coding test. Its GPT-6.1 Sol post says GPT-6.1 Sol "matches GPT-6 Astra at roughly one-fifth of the cost" on DeepSWE v1.1, at $2 / $10 per million tokens.

How each lab describes its model at work. Anthropic says Opus 5.5 "is particularly good at long and sprawling jobs like codebase-wide migrations and audits." OpenAI says that for coding, Astra "tends to be thorough in testing before considering a task complete," which on small tasks "can result in broader tests than the task requires," and that it is "more likely to ask for clarification where earlier models would make assumptions."

In our one-run test on November 16, both built a landing page with a sign-up form that fit a phone: Claude Opus 5.5 in 57 seconds for $0.15, and GPT-6 Astra in 65 seconds for $0.27. One run is one sample: 9 AI models, one landing page prompt.

Each lab's coding agent comes with its chat plans. ChatGPT Plus includes Astra in Codex, and OpenAI estimates 5 to 45 local Codex messages with Astra per 5 hours on Plus. Claude Pro includes Claude Code, and Anthropic's plans page lists Opus on Pro. How the 2 agents differ in practice: Codex vs Claude Code skills.

GPT-6 vs Claude Opus 5.5 price

What 3 typical requests cost at each model's Standard list rate, worked out from the pricing pages checked on November 16, 2026, with GPT-6.1 Sol for comparison:

RequestGPT-6 AstraClaude Opus 5.5GPT-6.1 Sol
100K input tokens, 10K output$1.50$0.60$0.30
The same, with the 100K input read from the prompt cache$0.60$0.22$0.11
500K input tokens, 20K output (past the 272K line)$11.50$2.40$2.30

The output side includes reasoning or thinking tokens, so the effort setting moves these totals. Writing to the cache the first time costs 1.25x the input rate on both: $12.50 per million for a 30-minute cache write on Astra, and $5 for a 5-minute write on Opus 5.5. GPT-6.1 Sol also charges more past 272K input tokens, $4 / $15, which is why it lands near Opus 5.5 in the last row.

Per task, the gap can close. OpenAI reports that at maximum effort on Terminal-Bench Science 0.1, a task costs $23.80 on Astra and $23.21 on Opus 5.5. Anthropic reports Opus 5.5 at its default effort matching Astra on Terminal-Bench 4.0 "for about 40% of the cost." Price per token is where to start; cost per finished task is what you pay.

Other rates from the same pages:

  • Batch: both halve the price. Astra costs $5 / $25 on Batch and Flex, and Opus 5.5 costs $2 / $10 on the Batch API.
  • Fast mode: Astra costs $20 / $100, 2x Standard. Opus 5.5 costs $8 / $40, for "up to 2.5x higher output tokens per second."
  • Where the model runs: OpenAI adds 10% for regional processing (data residency) on models released on or after March 5, 2026. Anthropic charges 1.1x for US-only inference.

On a subscription, you pay per month instead of per token. ChatGPT Plus ($20 a month) includes Astra in ChatGPT Work and Codex. In Chat, Astra powers "GPT-6 Pro," which comes with Pro (from $100 a month), Business, and Enterprise. Claude Pro ($20 a month, or $17 a month billed yearly) includes Opus and Claude Code. Max ($100) gives 5x Pro's usage, and Ultra ($200) 10x Pro's weekly usage.

GPT-6 vs GPT-5.6

GPT-5.6 came out on July 9, 2026, as 3 models, Sol, Terra, and Luna, and the gpt-5.6 alias points to GPT-5.6 Sol. GPT-6 followed with Astra on September 3, Sol and Luna on September 22, GPT-6.1 Sol on September 29, and GPT-6.1 Terra on October 26, per OpenAI's API changelog. So "GPT-6 vs GPT-5.6 Sol" can mean any of 3 GPT-6 models, here from OpenAI's model and pricing pages checked on November 16, 2026:

GPT-5.6 SolGPT-6 SolGPT-6.1 SolGPT-6 Astra
ReleasedJuly 9, 2026September 22, 2026September 29, 2026September 3, 2026
API price per 1M tokens$4 / $20, promotional until November 30, 2026$2 / $10$2 / $10$10 / $50
Cached input per 1M tokens$0.40$0.20$0.10$1
Context window1,050,0001,050,0001,050,0001,050,000
Knowledge cutoffFebruary 16, 2026April 20, 2026April 30, 2026April 30, 2026
Reasoning effortnone to max; default mediumnone to max; default mediumlow to max; default mediumlow to max; OpenAI doesn't say the default
OpenAI's description"GPT-5.6 flagship model for complex professional work""Built to power complex coding and agentic workflows""Near-Astra performance for complex work at a lower cost""Our most capable model for the most demanding work"

GPT-5.6 Sol's launch post listed it at $5 / $30; OpenAI's model page calls today's $4 / $20 "promotional pricing," available until November 30, 2026. GPT-6 Sol's model page points to GPT-6.1 Sol as "the newer Sol model." At $2 / $10, GPT-6.1 Sol costs half of GPT-5.6 Sol per token today.

What OpenAI's GPT-6 guide lists as new:

  • Async tool calling. The model keeps reasoning, or answers other parts of a request, while your code runs a tool.
  • Mid-turn steering. You send a correction while GPT-6 is working, and the Responses API keeps the work already done.
  • Changing reasoning effort mid-conversation without breaking the prompt cache, with a configuration_update input item.
  • Multi-agent and persisted reasoning on GPT-6.1 Sol. The model hands work to subagents and reuses reasoning from earlier turns.
  • Misalignment monitoring on GPT-6 Astra. OpenAI says that if the monitor pauses a task, ChatGPT and Codex may ask you to review the action; in the API, "the task will stop."

On behavior, OpenAI says: "GPT-6 Astra is generally better than GPT-5.6 Sol and earlier models at staying coherent during long tasks. It is also more likely to ask for clarification where earlier models would make assumptions." If you're moving code over, Astra and GPT-6.1 Sol drop the none reasoning effort ("use low instead"), and tool calls on them need the Responses API.

Benchmarks that cover both generations:

BenchmarkReported byGPT-5.6 SolGPT-6 Astra
Terminal-Bench 4.0 (agentic coding)Anthropic, September 22, as reported by OpenAI37.3%57.9%
FrontierCode v1.1, main set (agentic coding)Anthropic, September 2247.5%53.3%
Terminal-Bench-Science 0.1Anthropic, September 22 (Astra's figure as reported by OpenAI)22.4%64.6%
AutomationBench (business workflows)Anthropic, September 22, from Zapier's leaderboard28.8%41.4%
GDPval-AA v2.1 (knowledge work, Elo)Anthropic, September 2215881542
Cyber jailbreak evaluations, share refusedOpenAI, Path to Astra59%91.5%

OpenAI's GPT-6.1 Sol post, a week later, gives Astra 68.1% on Terminal-Bench Science 0.1.

In the ChatGPT app, OpenAI's help center says "GPT-6 replaces Latest" in the model menu, and "explicit GPT-5.6 Sol selections stay on GPT-5.6." With GPT-6 selected, GPT-6 Sol powers the Instant, Medium, High, and Extra High levels, and "Pro uses GPT-6 Pro, powered by GPT-6 Astra." Plus gets Medium and High; Extra High and Pro come with Pro, Business, and Enterprise. "The GPT-6 rollout does not retire older selectable models."

GPT-6 Astra vs Claude Fable 5.2

Claude Fable 5.2, released November 9, is Anthropic's top model for all users, and the closer match to GPT-6 Astra in price and purpose. Anthropic's November 9 table includes GPT-6 Astra. OpenAI's pages read for this guide don't mention Fable 5.2, so every benchmark in this section is Anthropic's. Both tables were checked on November 16, 2026.

GPT-6 AstraClaude Fable 5.2
ReleasedSeptember 3, 2026November 9, 2026
API price per 1M tokens$10 / $50$8 / $40
Prompts over 272K input tokens$20 / $75The same $8 / $40 across the full 1M
Cached input per 1M tokens$1$0.20
Fast mode$20 / $100$16 / $80
Context window / maximum output1,050,000 / 128,000 tokens1M / 128K tokens
Default effortOpenAI doesn't sayhigh
PlansChatGPT Plus (Work and Codex); Pro, from $100 (also in Chat)Claude Max, $100, and Ultra, $200, for up to half the weekly limit; Pro, $20, with usage credits
Benchmark (Anthropic, November 9)GPT-6 AstraClaude Fable 5.2Claude Opus 5.5
FrontierSWE v265.5%69.9%62.3%
FrontierCode 2.042.8%49.1% (xhigh effort)45.5%
Terminal-Bench-Science 0.168.1%74.6%58.7%
OSWorld 2.1, offline set (computer use)79.5%89.3%87.2%
Humanity's Last Exam, with tools57.2%74.2%67.7%
Remote Labor Index20.8%41.3% (high effort)21.3%
AA-Briefcase v1.1, Elo (knowledge work)156917851800

Anthropic notes that its table shows "each model's highest-available effort unless otherwise noted," with Fable 5.2 and Opus 5.5 run with safeguards on. On cost, it says that at high effort Fable 5.2 scores about the same as Opus 5.5's best on FrontierCode 2.0 "for roughly half the cost per task, and ahead of GPT-6 Astra's best score (42.8%) at about the same cost."

Anthropic doesn't send most work to Fable: "For most coding and knowledge tasks, we recommend that users start with Claude Opus 5.5, our strongest daily driver," and step up to Fable for "the most ambitious, open-ended technical work." Two limits come with Fable 5.2. Its safeguards "will block most cybersecurity-related tasks." It also "requires 30-day data retention and isn't available under zero data retention unless expressly authorized by Anthropic," while Anthropic says Opus 5.5 "is available with zero data retention." On Astra, OpenAI says access to its "most advanced cybersecurity capabilities will be more limited."

Which to use

If you wantStart withWhy, in the labs' words and prices
One model for most coding, at the lower price per tokenClaude Opus 5.5$4 / $20. Anthropic says it matches Astra on Terminal-Bench 4.0 "for about 40% of the cost"
OpenAI's models at a lower priceGPT-6.1 Sol$2 / $10. OpenAI: it "nearly matches GPT-6 Astra's intelligence on agentic coding, computer use, and professional work at one-fifth of Astra's standard input and output token prices"
Prompts over 272K tokens, such as a whole codebaseClaude Opus 5.5 for priceOne rate across 1M, where Astra charges 2x input and 1.5x output. On Google's GraphWalks long-context test (256K to 1M tokens), Astra scores 71.8% and Opus 5.5 66.8%
Scientific work in a terminalGPT-6 Astra, or Claude Fable 5.2OpenAI calls Astra's 68.1% on Terminal-Bench Science 0.1 "the highest score among the models tested." Anthropic's November 9 table gives Fable 5.2 74.6%
Computer useClaude Opus 5.5OSWorld 2.1: 87.2% to Astra's 79.5% in Anthropic's table, and 87.2% to 80.9% in Google's
The hardest, longest jobsClaude Fable 5.2 or GPT-6 AstraEach lab's top model: Fable 5.2 at $8 / $40, Astra at $10 / $50
An agent that asks before it assumesGPT-6 AstraOpenAI: "more likely to ask for clarification where earlier models would make assumptions"
A $20 planClaude Pro or ChatGPT PlusClaude Pro lists Opus and Claude Code. ChatGPT Plus has Astra in Work and Codex, and GPT-6 Sol in Chat

Before you switch a team or an app, run your own 10 hardest tasks through both at the same effort setting and compare the cost per finished task. OpenAI's own figures put the 2 at about $23 a task each on Terminal-Bench Science 0.1 at maximum effort, while their per-token prices are 2.5x apart.

If what you want from either model is a web page, such as a landing page with a sign-up form, ask for one HTML file: Convika publishes that file at a link and keeps the form's entries.

Frequently asked questions

Is GPT-6 better than Claude Opus 5.5 for coding?

Not on every test. On Anthropic's Terminal-Bench 4.0 run, Claude Opus 5.5 scores 66.4% and GPT-6 Astra 57.9%; on Google's run of the same test, 59.6% and 59.1%. GPT-6 Astra leads FrontierSWE v2, 65.5% to 62.3%, in both Anthropic's and Google's tables. Opus 5.5 costs $4 / $20 per million tokens, against Astra's $10 / $50.

How much does GPT-6 cost through the API?

GPT-6 Astra costs $10 per million input tokens and $50 per million output tokens, with cached input at $1. A prompt over 272K input tokens costs $20 / $75 for the whole request. The cheaper GPT-6 models cost $2 / $10 (GPT-6.1 Sol and GPT-6 Sol), $1 / $5 (GPT-6.1 Terra), and $0.10 / $0.50 (GPT-6 Luna). Batch and Flex halve these rates.

How much does Claude Opus 5.5 cost through the API?

$4 per million input tokens and $20 per million output tokens, with cache reads at $0.20 and 5-minute cache writes at $5. The rate stays the same across the full 1M-token context. The Batch API halves it, and fast mode costs $8 / $40.

Which is cheaper, GPT-6 or Claude?

Per token, Claude Opus 5.5 ($4 / $20) costs 40% of GPT-6 Astra ($10 / $50), and Claude Fable 5.2 ($8 / $40) costs 80%. OpenAI's GPT-6.1 Sol ($2 / $10) costs half of Opus 5.5. Per task, the gap can close: OpenAI reports $23.80 a task for Astra and $23.21 for Opus 5.5 on Terminal-Bench Science 0.1 at maximum effort.

What's the difference between GPT-6 and GPT-5.6?

GPT-6 is OpenAI's newer family: GPT-6 Astra came out on September 3, 2026, 8 weeks after GPT-5.6 on July 9. GPT-6.1 Sol costs $2 / $10 against GPT-5.6 Sol's promotional $4 / $20, has an April 30, 2026 knowledge cutoff instead of February 16, 2026, and adds multi-agent and persisted reasoning. OpenAI says Astra is better than GPT-5.6 Sol "at staying coherent during long tasks."

Is GPT-6 Sol the same as GPT-5.6 Sol?

No. GPT-6 Sol (September 22, 2026) is a GPT-6 model at $2 / $10 with an April 20, 2026 knowledge cutoff, and OpenAI points to GPT-6.1 Sol as "the newer Sol model." GPT-5.6 Sol (July 9, 2026) costs $4 / $20 until November 30, 2026, with a February 16, 2026 cutoff. In ChatGPT, GPT-6 in Chat runs on GPT-6 Sol, and choosing GPT-5.6 Sol keeps you on GPT-5.6.

Can I use GPT-6 Astra with ChatGPT Plus?

Yes, in ChatGPT Work and Codex. OpenAI's help center says "Plus plans include GPT-6 Astra in ChatGPT Work and Codex," and OpenAI estimates 5 to 45 local Codex messages with Astra per 5 hours on Plus. In Chat, Astra powers GPT-6 Pro, which comes with Pro (from $100 a month), Business, and Enterprise.

Does Claude Pro include Opus 5.5?

Yes. Anthropic's plans page lists Opus on Pro ($20 a month, or $17 billed yearly), Max ($100), and Ultra ($200), and Claude Opus 5.5 is Anthropic's newest Opus model. Pro also includes Claude Code. Fable comes with Pro through usage credits; Max and Ultra include it for up to half the weekly limit.

Which has the bigger context window, GPT-6 or Claude Opus 5.5?

About the same. GPT-6 Astra lists 1,050,000 tokens, up to 922,000 of them input, and Claude Opus 5.5 lists 1M tokens. Both cap one response at 128K output tokens. The difference is price: Astra charges 2x input and 1.5x output once a prompt passes 272K tokens, and Opus 5.5 keeps one rate.

GPT-6 Astra vs Claude Fable 5.2: which is better?

On Anthropic's November 9 table, Claude Fable 5.2 scores above GPT-6 Astra on every row that lists both, such as FrontierSWE v2 (69.9% to 65.5%) and Terminal-Bench-Science 0.1 (74.6% to 68.1%). OpenAI hasn't published a comparison with Fable 5.2. Fable 5.2 costs $8 / $40 per million tokens, against Astra's $10 / $50.

Why do the GPT-6 and Claude benchmark numbers differ between sources?

Each lab runs the test in its own setup. On Terminal-Bench 4.0, Anthropic ran Opus 5.5 at xhigh effort and quotes OpenAI's Astra figure at high, while Google's figures come from Artificial Analysis's runs. Anthropic also ran Opus 5.5 with production safeguards on and says "this likely reduces" its scores. Compare numbers within one table, not across tables.

Sources

Read on November 16, 2026, between 14:41 and 14:50 JST. OpenAI's GPT-6 Astra launch post showed "500 Internal Server Error," so OpenAI's own benchmark figures on this page come from its GPT-6.1 Sol post and Path to Astra page, and from Anthropic's footnoted quotes of OpenAI's figures.

Next step

How to password protect an HTML page or website

A JavaScript password script leaves the password in the page source. To password protect an HTML page or link, encrypt the file or let a server check it.

Related articles

More ways to send your page

Skipbox

Cheap & fast GitHub Actions runners, unlimited for $39 a month

Your own machine keeps Docker layers and package caches between jobs. Switch by changing one runs-on line.

Start a free Skipbox trial⁠

Share this article

Your model wrote a page. Put it at a link.

Upload the HTML file GPT-6 or Claude made, or paste its code. Convika publishes it, keeps the form's entries, and shows who visited.

Publish my AI-made page free

Free plan: no credit card, 1 published page, 50 form entries a month.