$ skilldirectory
— zsh — registry.skilldirectory.dev
skill libraryblogguidesabout

home / blog / research

ChatGPT GPT-6 Astra vs Claude Fable 5.2: Benchmarks, Pricing and Which to Use (2026)

Quick answer: On Anthropic's own launch tests, Claude Fable 5.2 beats GPT-6 Astra on every benchmark where both were measured, and its list price is 20% lower. But those numbers come from Anthropic, many gaps are small, and cost per task depends heavily on effort settings.

Run both on your own work before switching.

OpenAI's GPT-6 Astra (released September 3, 2026) and Anthropic's Claude Fable 5.2 (released November 9, 2026) are two of the most capable models you can pay for. This guide lays out the numbers, who published them, and where they are shaky.

Disclosure: Fable is made by Anthropic, the company behind Claude. Most of the head-to-head scores below come from Anthropic's own launch page, so they are self-reported. We label every source and flag where independent confirmation is missing.

What each model is

GPT-6 Astra (OpenAI). OpenAI calls it its most intelligent and aligned model. Per 9to5Mac's report, it started with a limited set of organizations and then widened to ChatGPT Plus, Pro, Business and Enterprise users, plus the API. Pro, Business and Enterprise users also get GPT-6 Astra Pro. Enterprise admins must switch it on for their workspace. OpenAI highlights computer use, multistep work like documents, spreadsheets and presentations, and software engineering.

Claude Fable 5.2 (Anthropic). Fable and Mythos are the same underlying model, according to Anthropic. Fable has safeguards that block high-risk cyber activity and most research biology tasks. Mythos 5.2 has fewer of those safeguards and is limited to vetted organizations through Anthropic's verification programs. Fable 5.2 is available on the Claude apps for eligible paid plans, in Claude Code, on the API as claude-fable-5-2, and on Amazon Web Services, Google Cloud and Microsoft Azure.

Benchmark chart

Bar chart comparing Claude Fable 5.2 and GPT-6 Astra on nine benchmarks from Anthropic's launch page
Scores in percent from Anthropic's Fable 5.2 launch page, November 9, 2026. Self-reported.
Benchmark (what it tests)Claude Fable 5.2GPT-6 Astra
FrontierCode 2.0 (agentic coding, Xhigh effort)49.1%42.8%
FrontierSWE v2 (agentic coding)69.9%65.5%
AutomationBench (business workflows)51.4%41.4%
Terminal-Bench-Science 0.174.6%68.1%
AA-Briefcase v1.1 (knowledge work, rubric)66.1%52.0%
Remote Labor Index (High effort)41.3%20.8%
Humanity's Last Exam, with tools74.2%57.2%
Chartography, no tools (chart reading)73.4%71.0%
RoboDojo Sim (robotics)64.9%29.8%

Two details worth knowing. On AutomationBench, Google's Gemini 4 Carbon scored 53.9%, above Fable. And on AA-Briefcase's head-to-head Elo rating, Claude Opus 5.5 (1800) edged Fable (1785). Anthropic reports Fable ahead of Astra in the table, but those are the results Anthropic chose to publish.

What OpenAI reports for Astra

OpenAI cited 98% on FrontierMath Tier 4, 99.9% on ARC-AGI-3 and 100% on ExploitBench, as reported by 9to5Mac. These are different tests from the ones in Anthropic's table, and we could not verify them, so they do not settle the comparison.

Independent index

Per Artificial Analysis figures cited by *The Batch*, GPT-6 Astra scored 53 on the Intelligence Index. We did not find a Fable 5.2 score from the same source, so we cannot compare them there.

Pricing

Bar chart of API prices: GPT-6 Astra $10 input and $50 output, Claude Fable 5.2 $8 and $40, Claude Opus 5.5 $4 and $20 per million tokens
API list prices in US dollars per 1 million tokens.
  • GPT-6 Astra API: $10 per million input tokens and $50 per million output tokens.
  • Claude Fable 5.2 API: $8 input and $40 output, down from $10 and $50 for Fable 5.1. Fast mode, up to 2.5x faster, costs $16 and $80. Batch pricing is $4 and $20, per VentureBeat.
  • ChatGPT plans: 9to5Mac reports Astra is included in existing subscription allowances, with credits for extra usage. It cites $20 a month for Plus and $100 or $200 a month for Pro.
  • Claude plans: Anthropic says Fable is included for up to 50% of weekly usage limits on Pro and standard Team seats, with a small number of resets depending on subscription length.

Anthropic recommends starting with Claude Opus 5.5 ($4 and $20) for most tasks and moving up to Fable only when needed. That advice from the maker is worth taking seriously, because a cheaper model that passes your tests beats a pricier one.

Cost per task is not the same as price per token

VentureBeat reports that Fable at High effort averaged $6.72 per task on FrontierCode 2.0, close to Astra at Max effort at $6.68, with Fable about 2.5 points higher. At Fable's Max setting the average was $44.70 per task and the score was lower than at Xhigh. Higher effort can cost a lot and may not help.

Caveats to keep in mind

  • Self-reported scores. Most head-to-head results come from Anthropic's launch materials. Independent runs may differ.
  • Different effort levels. Scores are tied to settings such as Xhigh or High, which change cost.
  • No uncertainty ranges. One partner result would shift slightly if timed-out runs counted as zero, per VentureBeat.
  • Safeguards limit some work. Fable blocks high-risk cyber activity and most research biology tasks, and sends some blocked requests to older models.
  • Different products. Astra is tied into ChatGPT, Codex and OpenAI's agent features, and Fable into Claude, Claude Code and cloud platforms. The surrounding tools matter as much as the model.

Which should you pick?

  • Choose Claude Fable 5.2 if your work is long-running coding or agentic tasks and you want Anthropic's reported lead, you already use Claude Code, or you want the model on AWS, Google Cloud or Azure.
  • Choose GPT-6 Astra if your team lives in ChatGPT and Codex and wants a single OpenAI stack.
  • Choose neither flagship if a cheaper model passes your tests. Both companies sell smaller models, such as Claude Opus 5.5 and OpenAI's GPT-6.1 Sol, that cover most work.

FAQ

Is Claude Fable 5.2 better than GPT-6 Astra? On Anthropic's published benchmarks, yes on all nine shown here. Those are self-reported, so treat the lead as promising, not proven.

Which is cheaper, Fable 5.2 or Astra? Fable's API list price is $8 and $40 per million tokens against Astra's $10 and $50. Real cost depends on how many tokens each uses at your effort setting.

What is the difference between Claude Fable and Claude Mythos? They are the same underlying model. Mythos has fewer cyber and biology safeguards and is limited to vetted organizations.

Can I use Fable 5.2 on a Claude subscription? Anthropic says Fable is included for up to 50% of weekly usage limits on Pro and standard Team seats, with details varying by plan.

Is GPT-6 Astra available in ChatGPT? Yes, per 9to5Mac, for Plus, Pro, Business and Enterprise users, with Enterprise admins needing to enable it.

Sources: Anthropic: Claude Fable 5.2 and Mythos 5.2, VentureBeat, 9to5Mac on GPT-6 Astra, The Batch.