Anthropic launches Claude Fable 5.2 at $8/$40, 20% below Fable 5.1

Anthropic released Claude Fable 5.2 and Claude Mythos 5.2 on November 9, 2026, and priced both at $8 per million input tokens and $40 per million output tokens on the Claude API. That is a 20% cut from Fable 5.1 and Mythos 5.1, which were listed at $10 and $50. The company describes the two releases as the same underlying model: Fable 5.2 ships with cybersecurity and biology safeguards and is open to every API customer, while Mythos 5.2 runs without those safeguards and is limited to organizations approved through Anthropic's Cyber and Life Sciences Verification Programs.

Both models keep a 1M-token context window, a 128K-token output limit and a June 2026 knowledge cutoff, and both are available from launch on the Claude API, Amazon Bedrock, Claude Models on Google Cloud, Microsoft Foundry and Claude Platform on AWS under the model ID claude-fable-5-2 (claude-mythos-5-2 for Mythos). Anthropic says output is up to 30% faster than the 5.1 generation, and an optional fast mode, priced at $16 per million input tokens and $80 per million output tokens, raises generation speed by up to 2.5x on selected platforms.

On Claude subscriptions, the launch comes with a time-limited offer. Pro subscribers and standard Team seats get all Fable models for one week, through November 16, 2026 at 11:59 pm PT, for up to 50% of their weekly usage limits, plus one Fable limit reset. Max, Ultra, premium Team seats and premium seats on legacy seat-based Enterprise plans receive between one and three Fable resets depending on how long they have been paying, valid until March 8, 2027.

··········

⁣⁣⁣⁣⁣⁣⁣⁣⁣⁣⁣⁣

CLAUDE FABLE 5.2 AND MYTHOS 5.2: ONE MODEL, TWO SAFEGUARD LEVELS.

Anthropic positions Fable 5.2 above Claude Opus 5.5 in its lineup. Its own guidance is to start most coding and knowledge work on Opus 5.5 and move up to Fable or Mythos for the hardest long-running tasks, or down to Sonnet 5.5 or Haiku 5.5 when cost and speed weigh more. The company also says Fable 5.2 can act as an advisor or orchestrator for cheaper executor models, a pattern the Claude API supports through its advisor tool, which is in beta.

The split between Fable and Mythos concerns safeguards, since the underlying capability is shared. On Fable 5.2, cyber and biology classifiers block high-risk requests such as penetration testing, exploit generation, binary-based vulnerability work and most research biology tasks. Blocked cyber queries are answered by Claude Opus 4.8 and blocked biology queries by Claude Opus 5, according to the announcement. Anthropic reports that Fable 5.2 triggers about 55% fewer blocks than Fable 5.1 in total, with roughly 60% fewer in cybersecurity and 40% fewer in biology, which should reduce the number of legitimate security and life-science requests that end in a fallback answer.

Mythos 5.2 removes those classifiers for vetted organizations. Mythos fast mode is also restricted to Verification Program members, and Enterprise customers using Claude Security now get Mythos 5.2 inside that product. Anthropic adds that it deliberately limited AI research and development capabilities during training, so performance on accelerator kernel optimization and language-model architecture work sits roughly at Opus 5.5 level, and some topics in that area fall back to Opus 5.

··········

CLAUDE FABLE 5.2 API PRICING COMPARED WITH FABLE 5.1 AND OPUS 5.5.

Every component of the Fable 5.2 price list is 20% lower than Fable 5.1, including prompt caching and the Batch API. Mythos 5.2 carries exactly the same prices as Fable 5.2. Prices below are in USD per 1M tokens, as listed on Anthropic's official pricing page on November 9, 2026.

........

Model	Input	Output	5-min cache write	1-hour cache write	Cache read	Batch input / output	Fast mode input / output
Claude Fable 5.1 (before)	$10	$50	$12.50	$20	$0.25	$5 / $25	Not listed
Claude Fable 5.2 (from Nov 9, 2026)	$8	$40	$10	$16	$0.20	$4 / $20	$16 / $80
Claude Mythos 5.2 (from Nov 9, 2026)	$8	$40	$10	$16	$0.20	$4 / $20	$16 / $80
Claude Opus 5.5 (reference)	$4	$20	$5	$8	$0.20	$2 / $10	$8 / $40

........

Fable 5.2 still costs twice as much as Opus 5.5 on base input and output, but the cache read price is now identical at $0.20, which narrows the gap for agent workloads that reread large, stable prompts. Anthropic's pricing page states that the full 1M-token window is billed at the standard per-token rate, so a very long request carries no surcharge. Fast mode, which Anthropic says speeds up token generation by up to 2.5x, doubles both rates to $16 and $80, cannot be combined with the Batch API, and stacks with caching multipliers. Requests pinned to US-only inference with the data residency setting pay a 1.1x multiplier on every category.

Fast mode for Fable 5.2 runs on the Claude API, in Claude Code and on both Google Cloud deployment modes, while Microsoft Foundry exposes it as a separate model, claude-fable-5-2-fast. Mythos 5.2 fast mode is offered on the Claude API to Verification Program members only.

··········

WHAT A CODING AGENT AND A BATCH JOB COST ON CLAUDE FABLE 5.2.

To show what the cut means in practice, two workloads were priced with the official rates. The first is an agentic coding session that reads 3 million input tokens over its lifetime, with 90% of those tokens served from the prompt cache and 10% billed at the base input rate, and that writes 150,000 output tokens. Cache write charges are left out to keep the comparison simple. The second is a document review job sent through the Batch API: 10,000 documents of 20,000 input tokens each, with 2,000 output tokens per document.

For the coding session on Fable 5.1, 300,000 uncached tokens at $10 cost $3.00, 2.7 million cached tokens at $0.25 cost $0.675 and 150,000 output tokens at $50 cost $7.50, for $11.175. On Fable 5.2 the same session costs $2.40 plus $0.54 plus $6.00, or $8.94. A team running 20 such sessions per working day over 22 working days would spend $4,917.00 a month on Fable 5.1 and $3,933.60 on Fable 5.2, saving $983.40.

........

Workload	Fable 5.1	Fable 5.2	Opus 5.5
One coding agent session (3M input, 90% cached, 150K output)	$11.18	$8.94	$4.74
440 sessions a month (20 a day, 22 days)	$4,917.00	$3,933.60	$2,085.60
Batch review of 10,000 documents (200M input, 20M output)	$1,500.00	$1,200.00	$600.00

........

The batch job falls from $1,500 to $1,200 because 200 million input tokens now cost $800 instead of $1,000 and 20 million output tokens cost $400 instead of $500. Output dominates the agent session, which is why the saving tracks the 20% headline closely there. The change favours long-running agents and heavy output workloads that already depend on Fable-class quality, while teams that can live with Opus 5.5 results still pay roughly half as much per session.

··········

CLAUDE FABLE 5.2 BENCHMARKS AGAINST OPUS 5.5, GPT-6 ASTRA AND GEMINI 4 CARBON.

All benchmark figures below come from Anthropic's launch page and have not been independently reproduced. Anthropic notes that cybersecurity scenario scores use different tasks for each lab, and that its internal robotics simulation ran at the highest available effort.

........

Benchmark	Fable 5.2	Fable 5.1	Opus 5.5	GPT-6 Astra	Gemini 4 Carbon
FrontierCode 2.0	49.1% (xhigh)	32.4% (xhigh)	45.5%	42.8%	41.1%
FrontierSWE v2	69.9%	56.3%	62.3%	65.5%	61.2%
AutomationBench	51.4%	31.4%	42.5%	41.4%	53.9%
Terminal-Bench-Science 0.1	74.6%	52.6%	58.7%	68.1%	64.8%
OSWorld 2.1 (offline, partial)	89.3%	83.5%	87.2%	79.5%	81.5%
Humanity's Last Exam (with tools)	74.2%	65.6%	67.7%	57.2%	Not reported
Remote Labor Index (high effort)	41.3%	17.9%	21.3%	20.8%	Not reported
AA-Briefcase v1.1 (Elo)	1785	1669	1800	1569	1611

........

The table shows Fable 5.2 ahead on most coding and agentic tests, with two exceptions in Anthropic's own data. Gemini 4 Carbon leads AutomationBench at 53.9% against 51.4%, and Opus 5.5 keeps a slightly higher AA-Briefcase Elo rating of 1800 against 1785, even though Fable 5.2 scores higher on the rubric version of that benchmark (66.1% against 61.2%). The largest jump over Fable 5.1 is on the Remote Labor Index, where the score more than doubles from 17.9% to 41.3%.

Anthropic also gives cost-adjusted comparisons. At high effort, Fable 5.2 scores 45.3% on FrontierCode, level with Opus 5.5's 45.5%, at roughly half the cost per task according to the company, and at low effort it matches Opus 5.5's best Humanity's Last Exam score for under a tenth of the cost. For Mythos 5.2, Anthropic reports 76.3% on its CyScenarioBench and 89.6% on CVE Rediscovery, where an October 2026 run over 1,829 CVEs found 10.2% more vulnerabilities than Opus 5.5 and 14.4% more than Mythos 5.1. The company further says a team of 1,000 Fable 5.2 agents completed a formalization of Fermat's Last Theorem in 30 hours, while Opus 5.5 in the same setup had reached 26% after 18 hours.

··········

[[ADPLUS_300x250_1]]

CLAUDE FABLE 5.2 ACCESS ON PRO, MAX, ULTRA, TEAM AND ENTERPRISE PLANS.

Outside the launch offer, Anthropic's plans page lists Fable on Pro as "paid with usage credits", includes Fable on Max ($100 a month) and Ultra ($200 a month) for up to half of the weekly limit, attaches it to premium Team seats for 50% of weekly limits, and marks it as available on Enterprise. The Free plan does not include Fable. The launch offers, described in Anthropic's help center, apply to Fable 5.2, Fable 5.1 and Fable 5 alike.

........

Plan or seat	Fable 5.2 launch offer	Offer expires
Pro and standard Team seats	All Fable models included for up to 50% of weekly limits, plus one Fable reset	Nov 16, 2026, 11:59 pm PT
Max, Ultra, premium Team, premium legacy Enterprise seats	1 reset (under 6 months paid), 2 resets (6 to 12 months), 3 resets (over 12 months)	Mar 8, 2027, 11:59 pm PT
Free, standard legacy Enterprise, self-serve and usage-based Enterprise	Not eligible	Not applicable

........

To qualify, paid billing must have been in place before 10:00 am PT on November 9, 2026; on Team and Enterprise, both the organization's billing and the member's join date must predate that moment, and subscriptions still in a free trial receive their resets once they convert to paid. Subscription length is counted from the first paid bill, or the first app store bill for app store subscribers, and cancelling and rejoining does not reset the clock. A reset restores the weekly usage spent on Fable models and fully refills the five-hour limit where an account still has one. Resets do not raise limits, cannot be undone, are lost on cancellation or downgrade and expire on the dates above. Fable resets are not offered on HIPAA-ready Enterprise plans.

Using a reset takes three steps in the Claude app: open Settings, then Usage, and select "Reset for free" under "Limit resets". The default model does not change afterwards, so Fable 5.2 has to be picked manually in the model selector. After November 16, Pro users who keep using Fable will draw on usage credits if they have enabled them. For the standing limits of each plan, see our guide to Claude usage limits across Free, Pro, Max, Team and Enterprise, and for the choice between the two top consumer tiers, our Claude Max vs Claude Ultra comparison.

··········

WHAT DEVELOPERS MUST CHANGE TO MIGRATE TO CLAUDE-FABLE-5-2.

Moving from Fable 5.1 starts with the model string, but Anthropic's documentation lists four breaking changes. The first concerns the thinking-block prefix check. If anything before a thinking block changes between turns, such as the system prompt, the tools array or an earlier message, Fable 5.2 returns a 400 error reading "The block is bound to a different conversation" on every account, whereas Fable 5.1 enforces this only for accounts created on or after August 31, 2026. Applications that keep system prompts and tools constant, only append messages and return assistant turns exactly as received are unaffected, as are requests built by Claude Code, claude.ai, Claude Managed Agents and the Claude Agent SDK. A temporary beta header lets the API drop mismatched blocks instead of failing, at the cost of extra output tokens.

The second change ties thinking blocks to the models that can read them. Fable 5.2 and Mythos 5.2 read blocks from each other and from the 5.5, 5.1 and 5 generations, but a router that switches from Fable 5.2 to Opus 5.5, Fable 5.1 or the Opus 4.8 and Opus 5 fallbacks silently loses that reasoning; the dropped blocks are not billed. Third, the computer_20251124 tool now returns a 400 error everywhere except Amazon Bedrock and must be replaced by computer_toolset_20260801. Fourth, a Fable 5.2 or Mythos 5.2 executor in the advisor tool beta accepts only a Fable 5.2 or Mythos 5.2 advisor, and the advisor's output now comes back encrypted.

Two request rules carry over from the 5.1 generation. Adaptive thinking is always on, so the enabled-with-budget and disabled thinking settings return errors, and forced tool use through tool_choice values "any" or a named tool is rejected; Anthropic recommends strict tool use or JSON outputs for schema-valid results. Effort runs from low to max with high as the API default, and the documentation advises starting at high and moving to xhigh only where a gain has been measured. Fable 5.2 requires 30-day data retention and is not offered under zero data retention unless Anthropic authorizes it, although the launch page says eligible customers can use it with zero retention ahead of adopting Enterprise Frontier Safeguards, which is in private beta. Anthropic commits to keeping the model available until at least November 9, 2027.

··········

⁣⁣⁣⁣⁣⁣⁣⁣⁣⁣⁣⁤

CLAUDE MYTHOS 5.2 SAFEGUARDS AND ALIGNMENT RESULTS.

Anthropic rates the risk of unsafe behavior by the new models in high-stakes situations as low. In its automated behavioral audit, where a lower score is better, Mythos 5.2 scored 2.67 against 2.79 for Opus 5.5, 3.10 for Sonnet 5.5 and 3.28 for Mythos 5.1. In a dedicated sandbox escape evaluation, Mythos 5.2 used outside credentials or files in 3.1% of scenarios, compared with 3.3% for Opus 5.5, 5.1% for Sonnet 5.5, 21% for Mythos 5.1 and 39% for Mythos 5. External testing included the UK AI Security Institute and METR.

The company also flags a limitation in its own evidence. Both new models voiced suspicion that they were being evaluated more often than recent predecessors, which Anthropic says limits how well test results predict behavior in real deployments. On the commercial side, Anthropic reports that anti-distillation techniques cut the data harvested by distillers by 90% to 99%, and that preserved thinking now applies to all accounts.

··········

CLAUDE FABLE 5.2: CONFIRMED FACTS AND OPEN QUESTIONS.

The verified part of the launch is commercial and technical. Anthropic's documentation confirms the November 9 release, the $8 and $40 base rates, the 20% reduction on every price component compared with Fable 5.1, the 1M-token context, the 128K output limit, the platform list and the four breaking API changes. The help center confirms the launch offers, their eligibility cutoff of 10:00 am PT on November 9 and their expiry dates of November 16, 2026 and March 8, 2027.

The performance claims are Anthropic's own and depend on effort settings that differ between rows, so independent results on coding and agent benchmarks will be the real test of whether Fable 5.2 justifies twice the per-token price of Opus 5.5. Anthropic has not published a retirement date for Fable 5.1 in its launch materials, and after November 16 Pro subscribers return to paying for Fable through usage credits, which makes the one-week inclusion a trial for that plan. For API buyers, the immediate result is a cheaper top tier with stricter conversation-integrity rules, which makes request-building code the first thing to audit before switching production traffic.

··········

FOLLOW US FOR MORE.

·····

DATA STUDIOS

·····

datastudios.org