Fable 5.2 is finally out, and the results are a lot to take in. The gap from 5.1 is what caught me first: 49.1% against 32.4% on FrontierCode 2.0, both at xHigh effort. On FrontierSWE v2, it goes from 56.3% to 69.9%. That’s a much bigger change than I’d expect from a small version-number bump. 1 5
There’s an independent result to go with it. Artificial Analysis puts Fable 5.2 Max at 65, ahead of Carbon High at 60, Opus 5.5 Max at 59 and Astra Max at 57 when I checked the leaderboard. Fable’s High and xHigh variants both display 62, so it leads there without having to run at Max. 2
The effort comparison is probably the part I’m most interested in trying. Anthropic says Fable 5.2 High reaches 45.3% on FrontierCode, almost level with Opus 5.5 Max at 45.5%, for roughly half the cost per task in its evaluation. At xHigh, Fable reaches that 49.1% result. Getting near Opus’s best result without paying for the highest setting would be useful. 1
It isn’t first in every row. Carbon still leads AutomationBench at 53.9% against Fable’s 51.4%, and Opus has the higher AA-Briefcase Elo. The table also mixes in clearly marked Mythos 5.2 results for cybersecurity and protein design. Fable and Mythos share an underlying model, but have different safeguards and access rules, so those rows shouldn’t all be read as scores for the Fable you can open in Claude. 1
A week to try it on Pro
The other welcome bit is access. Pro subscribers and standard Team seats can use Fable within their plan through 16 November, 11:59 pm PT, up to 50% of their weekly usage limit. Eligible accounts also get a Fable reset. It’s a launch offer, so it doesn’t mean permanent unlimited access: after the week, or after that allowance is used, Fable runs on usage credits if they’re enabled. 3 6
For API users, Fable 5.2 costs $8 per million input tokens and $40 per million output tokens. Anthropic says output generation is up to 30% faster than 5.1. Optional fast mode costs $16/$80 and can generate tokens up to 2.5 times faster; that’s a token-generation claim, rather than a promise that every task finishes 2.5 times sooner. The API identifier is claude-fable-5-2. 1
We’ll be adding Fable 5.2 to our benchmarks soon. For now, the launch numbers give me plenty to be excited about, especially seeing High get so close to Opus Max in that coding comparison. Our first Low-effort seaplane clip is already in Demos; I’ll give the hands-on results their own article once we’ve done more.
Sources & references
- Anthropic: Introducing Claude Fable 5.2 and Claude Mythos 5.2 ↗
9 November 2026. Official launch, API pricing, efficiency claims and evaluation table. Results use highest-available effort unless noted; several specialist rows explicitly refer to Mythos rather than Fable.
- Artificial Analysis: LLM leaderboard ↗
Public leaderboard checked on 9 November 2026. Displayed scores: Fable 5.2 Max 65, Carbon High 60, Opus 5.5 Max 59 and Astra Max 57. The supplied chart is retained as a launch-day snapshot, not a Lumina ranking.
- Claude Help Center: Fable 5.2 launch offers ↗
Official terms checked on 9 November 2026: Pro and standard Team seats receive included Fable access up to 50% of weekly limits through 16 November, 11:59 pm PT. Reset eligibility has a paid-billing cutoff; ongoing access afterwards uses usage credits.
- Lumina: Fable 5.2 release post ↗
9 November 2026. Our release finding and supplied discovery card.
- Lumina: reaction to the Fable 5.2 results ↗
9 November 2026. Our reaction to the published benchmark table.
- Lumina: Fable’s launch-week Pro access ↗
9 November 2026. Our post on the temporary Pro offer; eligibility and limits are taken from the official help page.
