Claude Fable 5.2 Is Out: Price, Benchmarks & GPT-6.1

By Versus Tools AI Team · Published · Updated · 13 min read · AI Models & News

We may earn a commission if you buy through links on this page. Read the affiliate disclosure · How we review tools

Quick answer. Claude Fable 5.2 and Claude Mythos 5.2 were released Nov 9, 2026. They are the same model: Fable is the general release at $8 per million input tokens and $40 per million output tokens, and Mythos is for Anthropic's verification programs. Artificial Analysis lists GPT-6.1 Terra at $1 and $5. We did not test it.

Claude Fable 5.2 is out. Anthropic released it on Nov 9, 2026, together with Claude Mythos 5.2, and the two names are one underlying model. The general release is claude-fable-5-2 at $8 per million input tokens and $40 per million output tokens, with a 1 million-token context and 128K max output. Mythos 5.2 uses the same spec and the same list price, and access is limited to Anthropic’s verification programs.

This page is based on official documentation, benchmarks and public reports. We did not run Claude Fable 5.2, Claude Mythos 5.2, or the leaderboards below.

How this page was checked

We read Anthropic’s Nov 9, 2026 announcement, the Claude API release notes for that day, the Fable 5.2 model overview, and Anthropic’s pricing page. Rival token prices come from OpenAI’s GPT-6.1 Terra post, the Gemini Developer API pricing page, and the write-ups already on this site for Gemini 4 Hydrogen and Qwen4-Max. The Artificial Analysis comparison page is used only for the figures it prints, and those cells are marked as Artificial Analysis. If a vendor page does not print a cell, the table says “not published”.

Google’s frontier model from the next day is Gemini 4 Carbon. Both launches are on the AI model release timeline and in the best AI models ranking.

Key facts

Claude Fable 5.2Claude Mythos 5.2
ReleasedNov 9, 2026Nov 9, 2026, same announcement
MakerAnthropicAnthropic. Same underlying model as Fable 5.2
API model idclaude-fable-5-2claude-mythos-5-2
Who can call itGeneral release. The announcement says it is on Claude’s platforms, with extra Fable usage on eligible paid plans this weekOrganizations in Anthropic’s verification programs, including Cyber and Life Sciences
Context1 million tokensSame, per the model overview
Max output128K tokensSame
Input and outputText and images in. Text outSame capabilities, per Anthropic
ThinkingAdaptive, always on. Cannot be turned offSame
Default efforthigh on the Claude API. The announcement says Medium is the default in Claude apps and Claude CodeSame underlying model. The overview does not print a separate Mythos default
Knowledge cutoffJun 2026, listed as both the reliable cutoff and the training-data cutoffNot given a separate date. Anthropic says the specifications are shared
List price$8 / 1M input, $40 / 1M outputSame list price
CacheRead $0.20. 5-minute write $10. 1-hour write $16. Reads are 0.025x the base input priceSame
Fast mode$16 / $80. Up to 2.5x output tokens per secondSame list price. The announcement limits Mythos fast mode to verification-program organizations
Batch API$4 / $20, a 50% discount. Fast mode is not available with BatchSame batch rates on the pricing page
Data retention30 days, unless Anthropic expressly authorizes zero data retentionSame
RetirementNot sooner than Nov 9, 2027The overview states the shared spec on the Fable 5.2 page. A separate Mythos retirement date is not printed there

What changed from Claude Fable 5.1

Fable 5.1 and Mythos 5.1 launched on Sept 1, 2026, at $10 per million input tokens and $50 per million output tokens, with cache reads at $0.25. Fable 5.2 keeps the 1 million-token context, the 128K max output, and always-on adaptive thinking, and the list price is lower.

Anthropic says the new pair generates outputs up to 30% faster than Fable 5.1 and Mythos 5.1, and that an optional fast mode speeds token generation by up to 2.5 times. On Medium effort, which Anthropic calls the default in Claude apps and Claude Code, the announcement says Fable 5.2 can finish challenging tasks with fewer tokens than Fable 5.1. On the API, the model overview’s default effort is high, so the app default and the API default are not the same setting.

Anthropic’s recommendation for routing is specific. For most coding and knowledge work, start with Claude Opus 5.5, then move up to Fable or Mythos when the task needs it, or down to Sonnet 5.5 or Haiku 5.5. Opus 5.5 is $4 / $20 on the same pricing page. Fable can also sit as an advisor that smaller models consult, or as an orchestrator that hands simpler work to those models. The Nov 9 release notes say a Fable 5.2 or Mythos 5.2 executor accepts only Fable 5.2 or Mythos 5.2 as its advisor, and that every executor model accepts Fable 5.2 as an advisor.

The safeguards are the practical difference between the two Nov 9 names. Fable blocks high-risk cyber tasks, including penetration testing, exploit generation, and binary vulnerability work, and it blocks most research-biology tasks. When a query is blocked, Anthropic says the user gets a response from Opus 4.8 for cybersecurity and from Opus 5 for biology. Early testing on claude.ai, Claude Code, and Cowork showed about 55% fewer blocks than Fable 5.1, about 60% fewer in cybersecurity and 40% fewer in biology. Mythos is the version for vetted cybersecurity and life-sciences organizations.

Claude Fable 5.2 pricing

Checked on Anthropic’s pricing page and the Fable 5.2 model overview on Nov 16, 2026. Prices are USD per million tokens.

StandardFast modeBatch API
Input$8$16$4
Output$40$80$20
5-minute cache write$10Cache multipliers apply on top of fast mode. A separate fast-mode cache card is not printed in the section we usedThe batch table prints input and output only
1-hour cache write$16Same noteSame note
Cache read$0.20Same noteSame note

Fast mode is set with speed: "fast" on Messages API requests, with no beta header, according to the Nov 9 release notes. The pricing page says this listed fast-mode rate is the Claude API price, and that Google Cloud sets its own price on Claude Models on Google Cloud (Google managed). Do not assume the Google-managed deployment is $16 / $80.

Artificial Analysis prints $8 input and $40 output for Fable 5.2, and $1 and $5 for GPT-6.1 Terra. That pair matches the two vendors’ own pages. Its cache-hit column is $0.20 for Fable, which matches Anthropic. Its cache-write column is $10, which matches Anthropic’s 5-minute write and not the 1-hour write of $16. For Terra, Artificial Analysis prints a $1.25 cache write. OpenAI’s launch post prints cached input at $0.05 and does not print a cache-write price. The model page, read Nov 17, 2026, does print cache writes at $1.25 for 30-minute retention, which matches that Artificial Analysis column. The 12-hour write on the same page is $2.50.

Benchmarks Anthropic published

Every score in this section is from Anthropic’s Nov 9 post. We did not rerun it. Anthropic says scores for Fable 5.2, Fable 5.1, and Opus 5.5 use safeguards unless a cell says otherwise, and that a cell is the model’s highest-available effort unless the cell names an effort level. GPT-6 Astra in this table is Anthropic’s column. It is not GPT-6.1 Terra, and OpenAI’s Terra post is a different set of numbers, covered in the next section.

Fable 5.2Fable 5.1Opus 5.5GPT-6 AstraGemini 4 Carbon
FrontierCode 2.049.1% at Xhigh32.4% at Xhigh45.5%42.8%41.1% at Medium
FrontierSWE v269.9%56.3%62.3%65.5%61.2%
AutomationBench51.4%31.4%42.5%41.4%53.9%
Terminal-Bench-Science 0.174.6%. Mythos 5.2 is 75.3%52.6%58.7%68.1%64.8%
OSWorld 2.1, offline set, partial89.3%83.5%87.2%79.5%81.5%
Chartography, no tools73.4%44.8%64.4%71.0%72.5%
AA-Briefcase v1.1, rubric66.1%61.5%61.2%52.0%61.8%
AA-Briefcase v1.1, Elo17851669180015691611
Humanity’s Last Exam, with tools74.2%65.6%67.7%57.2%not in the table
Remote Labor Index41.3% at High17.9%21.3%20.8%not in the table
WANDR, relaxed score82.1%68.7%72.3%not in the tablenot in the table
RoboDojo Sim64.9%27.6%39.4%29.8%not in the table

Two cells are easy to misread. On FrontierCode 2.0, the chart text says Fable 5.2 at High effort scores 45.3%, about level with Opus 5.5’s best score of 45.5%, and the 49.1% figure is Xhigh. On AA-Briefcase, Fable’s rubric is higher than Opus 5.5 and its Elo is lower, 1785 against 1800. Carbon’s AutomationBench cell, 53.9%, is higher than Fable’s 51.4% in this table. Google’s own Carbon post says Carbon ranks number 1 on Zapier’s AutomationBench and does not print the score. This page does not treat those as one confirmed number.

Anthropic also describes results that are not a public shared leaderboard. A chart note puts BenchCAD 2.1 at 83.5 for Fable 5.2 at Max effort, against 65.3 for Opus 5.5 and 44.4 for Fable 5.1. On CyScenarioBench, Mythos 5.2 is listed at 76.3%, and Anthropic says each lab’s scores on that test use unique tasks. A 1,000-agent formalization of Fermat’s Last Theorem is described as finishing in 30 hours, against Opus 5.5 reaching 26% in 18 hours in the same setup, one run per model. Those are Anthropic’s accounts.

Claude Fable 5.2 vs GPT-6.1 Terra

The dedicated head-to-head, with the Nov 17, 2026 model-page specs and the Artificial Analysis rows re-read that day, is Claude Fable 5.2 vs GPT-6.1 Terra.

OpenAI’s GPT-6.1 Terra announcement does not print a calendar date. OpenAI’s research release index dates “Introducing GPT-6.1 Terra” to Oct 26, 2026. The API id in the announcement is gpt-6.1-terra. Standard API pricing there is $1 per million input tokens, $0.05 per million cached input tokens, and $5 per million output tokens. The announcement does not print a context window. The model page, read Nov 17, 2026, does: a 1,050,000 context window, 922,000 maximum input tokens, and 128,000 max output tokens. It also lists cache writes at $1.25 per million tokens for 30-minute retention and $2.50 for 12-hour retention.

OpenAI’s own coding figure, at maximum reasoning, is 50.3% on Terminal-Bench 4.0, against 21.5% for GPT-5.6 Terra. OpenAI says GPT-6.1 Sol and GPT-6 Astra remain the better choice for the most complex coding tasks. That 50.3% is OpenAI’s number.

Artificial Analysis, on its Fable 5.2 versus GPT-6.1 Terra comparison page, prints a different Terminal-Bench 4.0 pair and an Intelligence Index. These are not OpenAI’s or Anthropic’s scores. The page’s Intelligence Index is version 5.0.1.

Artificial AnalysisClaude Fable 5.2GPT-6.1 Terra
Input / output per 1M tokens$8 / $40 at every effort row$1 / $5 at every effort row
Cache hit / cache write per 1M$0.20 / $10$0.05 / $1.25
Context on that page1M1M
Intelligence Index, max6551
Intelligence Index, high6242
Terminal-Bench 4.0, max59.6%55.6%
Terminal-Bench 4.0, xhigh61.6%46.5%
Cost per Intelligence Index task, low$3.10$0.0713
Cost per Intelligence Index task, max$26.54$0.96
Time per task, low5.5 minutes1.2 minutes

On that page, Fable’s xhigh Terminal-Bench 4.0 score is higher than its max score. Max effort is not the top cell in that row. Output speed on the same page sits around 102 to 109 tokens per second for the Fable rows and around 64 to 72 for the Terra rows. The $1.25 Terra cache write matches the model page’s 30-minute rate, read Nov 17, 2026. The launch post still does not print it.

Price and spec comparison

“Not published” means the official page we used for that model did not print the cell. Qwen and Hydrogen figures are the ones already sourced on their pages on this site.

Fable 5.2Fable 5.1Opus 5.5GPT-6.1 TerraGemini 4 CarbonQwen4-Max
ReleasedNov 9, 2026Sept 1, 2026not published on the Fable 5.2 pagesOct 26, 2026 on OpenAI’s research indexNov 10, 2026Nov 16, 2026
API idclaude-fable-5-2claude-fable-5-1not repeated heregpt-6.1-terragemini-4-carbonqwen4-max
Input / 1M$8$10$4$1$2 through Nov 30, 2026, then $4$1.50 in Singapore
Output / 1M$40$50$20$5$10 through Nov 30, 2026, then $20$4.50 in Singapore
Cache-hit input / 1M$0.20$0.25$0.20$0.05 cached input, on OpenAI’s post$0.10, then $0.20. Implicit only$0.075 in Singapore
Context1M1M1M1,050,000 on the model page, read Nov 17, 2026. 1M on Artificial Analysis1,048,576 combined input and output on Google Cloud’s model page1,000,000
Max output128K128K128K128,000 on the model page, read Nov 17, 2026Long decode up to the rest of the 1,048,576 window. See Gemini 4 Carbon131,072 non-thinking

Gemini 4 Hydrogen is the cheaper Google coding model, at $1.50 / $7.50, and it has a free API tier. Carbon does not. Hydrogen’s Terminal-Bench number is Google’s 4.1 score and is not comparable with the 4.0 scores above.

👍 Pros

  • List price is $8 / $40, down from $10 / $50 on Fable 5.1, with the same 1M context and 128K max output
  • Anthropic's table shows gains over Fable 5.1 on every shared row we copied, and cache reads are $0.20
  • Documented on the Claude API, Amazon Bedrock, Google Cloud, and Microsoft Foundry

👎 Cons

  • $8 / $40 is several times the GPT-6.1 Terra and Gemini 4 Carbon list prices
  • Mythos, not Fable, is the version for the cyber and biology work Fable's classifiers block
  • The headline benchmarks are Anthropic's or Artificial Analysis's. We did not reproduce them

Who should upgrade

Upgrade from Fable 5.1 or Mythos 5.1 if you already pay for that tier. The new ids are cheaper on the list card, Anthropic reports up to 30% faster generation before fast mode, and the benchmark table above is a step up from 5.1 on the rows Anthropic printed. Teams that were hitting safeguard blocks may also see fewer of them. Anthropic’s early figure is about 55% fewer blocks than Fable 5.1.

Stay on Opus 5.5 for most coding and knowledge work. That is Anthropic’s own instruction, and Opus 5.5 is $4 / $20. Move to Fable 5.2 when the task is the open-ended technical work the announcement is about: long agent sessions, large migrations, and research that smaller models are not finishing.

Use Mythos 5.2 only if you need the capability Fable blocks and you can pass the Cyber Verification Program or the Life Sciences Verification Program. Claude Enterprise customers can also use Claude Security, which Anthropic says now includes Mythos 5.2, without that being the same thing as open Mythos API access.

Choose a different lab when the bill is the constraint. GPT-6.1 Terra is $1 / $5 on OpenAI’s post. Gemini 4 Carbon is $2 / $10 through Nov 30, 2026. Qwen4-Max is $1.50 / $4.50 in Singapore. Artificial Analysis’s cost-per-task column is one illustration of the gap versus Terra, at $3.10 against $0.0713 on its low setting. None of those prices is a claim that the cheaper model won a test we ran.

How to access it

Anthropic’s availability table on the model overview:

  • Claude API and Claude Platform on AWS: claude-fable-5-2.
  • Amazon Bedrock (bedrock-runtime): global.anthropic.claude-fable-5-2.
  • Claude Models on Google Cloud, Google managed and Anthropic managed: claude-fable-5-2.
  • Microsoft Foundry: claude-fable-5-2.
  • Claude apps: the announcement says Fable 5.2 is available on all of Anthropic’s platforms. For this launch week, plans and seats that include Fable models get one to three Fable resets based on how long they have been subscribed, and Fable is included for up to 50% of weekly usage limits on Pro plans and on standard Team seats.
  • Fast mode: the announcement lists Claude Code, the Claude Platform, both Google Cloud Claude deployments, and Microsoft Foundry (Anthropic hosted). Mythos fast mode stays inside the verification programs.
  • Mythos API id: claude-mythos-5-2, via the verification programs. The release notes name the Cyber Verification Program as an example. Life sciences applicants use the Life Sciences Verification Program.

We are not publishing a sample request. Confirm the id for your platform in the console you actually bill, especially on Google-managed Claude, where Anthropic says Google sets the fast-mode price.

The pages we read do not name a mainland China region, a China app, or an official third-party API router. Access as documented is the Claude surface and the cloud platforms above. This article does not add an unofficial route.

See Cursor →

Fable 5.2 is a model. These products are separate, and a commission does not change the prices above.

  • Cursor if the work is the editor around a coding model.
  • Writesonic if you want a writing product rather than the Claude API.
  • Fireflies for meeting notes, which the model will not record for you.

Sources

Read on Nov 16, 2026.

Pricing and availability last checked Nov 17, 2026. The Terra context window and cache-write rate are from the model page read that day. This page contains affiliate links. A commission does not change the model facts.

Frequently asked questions

When was Claude Fable 5.2 released?

Anthropic's announcement is dated November 9, 2026. The Claude API release notes for that day list claude-fable-5-2 and claude-mythos-5-2. The model overview says Fable 5.2 will not be retired sooner than November 9, 2027.

What is Claude Fable 5.2 pricing?

Anthropic's pricing page lists $8 per million input tokens and $40 per million output tokens. Cache reads are $0.20, 5-minute cache writes are $10, and 1-hour cache writes are $16, per million tokens. Fast mode is $16 and $80. The Batch API is $4 and $20. Claude Fable 5.1 is $10 and $50 on the same page.

What is Claude Mythos 5.2?

Anthropic says Claude Mythos 5.2 is the same underlying model as Claude Fable 5.2, at the same price, with the same 1 million-token context and 128K max output. Mythos is limited to organizations in Anthropic's verification programs, including the Cyber Verification Program and the Life Sciences Verification Program. Fable is the general release and keeps the cyber and biology safeguards.

How does Claude Fable 5.2 vs GPT-6.1 compare on price?

On the pages checked Nov 16, 2026, Claude Fable 5.2 is $8 per million input tokens and $40 per million output tokens. OpenAI lists GPT-6.1 Terra at $1, $0.05 cached input, and $5. Artificial Analysis prints the same $8/$40 and $1/$5 pair. A lower token price is not a quality ranking.

Where can I use Claude Fable 5.2, including from China?

Anthropic's model overview lists the Claude API, Amazon Bedrock, Claude Platform on AWS, Claude Models on Google Cloud (Google managed and Anthropic managed), and Microsoft Foundry. The announcement also describes eligible paid Claude plans. Those pages do not name a mainland China endpoint or a third-party API router.

Did you test Claude Fable 5.2?

No. This page is based on official documentation, benchmarks and public reports read on Nov 16, 2026. Scores from Anthropic and from Artificial Analysis are labeled with the page they came from.