Claude Fable 5.2 Is Now on Metir: When to Pick It (2026)

Anthropic released Claude Fable 5.2 on November 9, 2026, and it is now in the Metir model picker on web and mobile, replacing Fable 5.1. This post covers the practical side of that change: what it means for your chats, when Claude Fable 5.2 is the right pick and when a cheaper Claude model is, and the API-level changes that shaped the upgrade. For the benchmark table, the safeguard details and the Mythos 5.2 story, see our full news analysis of the release.

$8 / $40per million input / output tokensdown from $10 / $50 on Fable 5.1

74.6%Terminal-Bench-Science 0.1vs 52.6% for Fable 5.1

45.3%FrontierCode 2.0 at High effortOpus 5.5's best is 45.5%

30 daysdata retention requiredno zero data retention by default

Anthropic

Microsoft

Fable 5.2 runs on Metir directly through Anthropic and through Microsoft Foundry.

What changes for Metir users

The change is small by design. Claude Fable 5.2 appears in the model picker on web and mobile, both as the direct Anthropic model and through Microsoft Foundry. Fable 5.1 is replaced by it.

No action needed. Anyone who already had Fable selected is moved to 5.2 automatically, including existing chats set to the Foundry variant.
Still a deliberate choice. Fable stays picker-only. The auto model router never selects it, because it is the highest-priced of the current Claude models Anthropic lists and most requests do not need it.

That picker-only posture matches how Anthropic describes the model. Its model documentation positions Fable 5.2 for "demanding reasoning and long-horizon agentic work", such as agent sessions that run for hours, multistep deep research, and analysis carried through to a finished document, spreadsheet or deck.

Which Claude to pick: Opus 5.5, Sonnet 5.5 or Fable 5.2

Anthropic's own guidance is to start lower. Its documentation keeps Claude Opus 5.5 as the default for most workloads, and the what's new page is specific about the step up: move to Fable 5.2 when your evals on an Opus model at higher effort still do not meet your requirements. The launch post frames Fable and Mythos as best for "the most ambitious, open-ended technical work."

If your task is...	Start with	Why
Everyday drafting, analysis, coding help	Opus 5.5	Anthropic's recommended starting point, at $4 / $20 per million tokens
High-volume or latency-sensitive work	Sonnet 5.5	Listed as fast, at $2 / $10 per million tokens
Hours-long agent runs, deep research, hard multistep reasoning	Fable 5.2	The model Anthropic positions for this work, at $8 / $40
Opus 5.5 at high effort still falls short	Fable 5.2	The step-up condition in Anthropic's guidance

Anthropic list prices: where Fable 5.2 sits in the Claude lineup

USD per million tokens at standard rates, from Anthropic’s model documentation. Batch processing is 50% off and fast mode is priced higher, so neither is shown.

Fable 5.2 is 20% below Fable 5.1 on both rates, and still about twice the price of Opus 5.5.

Effort is the main dial

Fable 5.2 supports five effort levels (low, medium, high, xhigh, max) and defaults to high on the API. Anthropic says most of the capability gain is already present at high, and suggests xhigh only for the hardest long-running coding and agentic work where a gain has been measured.

Anthropic also claims the effort ladder changes the cost comparison. On FrontierCode 2.0, Fable 5.2 at High effort scores 45.3%, about level with Opus 5.5's best of 45.5%, at roughly half the cost per task. Those are vendor-reported figures, and on a different task mix the result can differ, so the practical advice is the same as Anthropic's: run your own evals at more than one effort level.

The cheaper model is the right default until your own evals say otherwise.
On Anthropic's tiered model guidance

On raw capability, the headline gap is widest on long-horizon agentic work. Anthropic reports 74.6% on Terminal-Bench-Science 0.1 for Fable 5.2 against 52.6% for Fable 5.1 and 58.7% for Opus 5.5. All figures are self-reported, and the full comparison includes rows where Fable 5.2 does not lead.

Two men standing in the lobby of Anthropic's San Francisco office, with wood-panelled seating booths behind them — A lobby at Anthropic's San Francisco office during an October 2024 visit by the UK Secretary of State for Science, Innovation and Technology. The photo predates Fable 5.2 and shows the office only. Photo by the Department for Science, Innovation and Technology via Wikimedia Commons, CC BY 2.0.

The API changes that shaped the upgrade

Fable 5.2 is mostly a model-ID swap from 5.1 to claude-fable-5-2, but Anthropic lists several changes that can break code. These are the ones most relevant to a product that lets people choose among models.

A lower list price

Fable 5.2 costs $8 per million input tokens and $40 per million output tokens, against $10 and $50 for Fable 5.1, a 20% reduction on both rates. Cache reads cost $0.20 per million tokens, and batch processing is 50% off. Anthropic also offers an optional fast mode at $16 and $80. These are Anthropic's list prices, shown here for comparison between models.

Preserved thinking is enforced on every account

Every thinking block records the model that produced it, and the API checks whether anything before that block has changed. On Fable 5.2, a request that sends back an invalid thinking block gets a 400 error on any account. On Fable 5.1 the same error applied only to accounts created on or after August 31, 2026, so code that edits its own history could pass on an older account and fail after switching. Anthropic's docs describe a beta setting that drops affected blocks instead, but call it a stopgap.

Thinking blocks do not travel to other models

Only Fable 5.2 and Mythos 5.2 read Fable 5.2 thinking blocks. If a conversation moves to another model, including Opus 5.5 or Fable 5.1, the reasoning from the Fable turns is dropped before the next model sees it. The request still succeeds, and dropped blocks are not billed. For anyone who switches models partway through a long task, that is a reason to finish a reasoning-heavy stretch on one model.

Forced tool choice is still rejected

Setting tool_choice to any or to a named tool returns a 400 error on Fable 5.2. For schema-valid JSON, Anthropic points to strict tool use or JSON outputs instead. Adaptive thinking is always on and cannot be disabled, and sampling parameters such as temperature must be omitted.

30-day retention

Fable 5.2 requires 30-day data retention and is not available under zero data retention unless Anthropic expressly authorizes it. That matters for regulated teams whose policies require zero retention.

Behaviour worth knowing

Anthropic notes that a prompt line asking for less thinking does not stop Fable 5.2 from thinking at medium effort and above, and that on long tasks a turn can end before the work is done, more often at low and medium effort. The fix it suggests for the first is a lower effort level, and for the second is a prompt that says to finish the whole task.

Where this leaves the choice

Fable 5.2 is cheaper than Fable 5.1 and, on Anthropic's numbers, more capable on long-running agentic work. It is still the highest-priced model in Anthropic's current lineup, and Anthropic itself tells most users to start with Opus 5.5. The sensible use of a picker is to match the model to the task: a cheaper model for routine work, Fable when the task is long, hard and worth the extra spend.

Sources:

Image credits

Header image: Anthropic co-founder Jack Clark giving a talk on AI at the Schwarzman Centre in Oxford, May 2026, by Mvolz via Wikimedia Commons, released under CC0. In-body photo: Secretary of State Peter Kyle visits Anthropic in San Francisco, 25 October 2024, by Alecsandra Dragoi / Department for Science, Innovation and Technology via Wikimedia Commons, licensed under CC BY 2.0.