Claude Fable 5.2 Cuts Its Price 20% and Tops the Intelligence Index. At Max Effort, a Task Costs $26.54, 2x Fable 5.1.

$8 and $40: what changed on the price sheet

Fable 5.2 ships alongside Claude Mythos 5.2. Anthropic says they are the same underlying model: Fable carries cybersecurity and biology safeguards and is open to everyone, Mythos goes to organizations in Anthropic’s Cyber and Life Sciences Verification Programs.

Every Fable rate moved down by the same 20%. Anthropic’s announcement just says “a lower price”; the 20% comes from its own pricing page, which still lists Fable 5.1.

Anthropic’s pricing and models pages; OpenAI’s GPT-6.1 Sol docs. Dashes: not listed for that model. GPT-6.1 Sol charges 2x input and 1.5x output on prompts over 272K tokens; Anthropic’s current models bill the full 1M context at standard rates.

Two details matter for builders. Fable 5.2 costs exactly 2x Opus 5.5 per token, and Opus 5.5 in fast mode costs the same $8 / $40 as Fable 5.2 in standard mode. The model ID is claude-fable-5-2. Anthropic’s models page lists its API default effort as high; in Claude apps and Claude Code the default is medium.

On speed, Anthropic claims output “up to 30% faster” than Fable 5.1. Artificial Analysis measures 103 output tokens per second for Fable 5.2 at max effort and 87 for Fable 5.1, an 18% gain on its runs. Fast mode promises up to 2.5x on top, at 2x the price.

Anthropic’s table: 10 wins, 2 losses

Anthropic compared Fable 5.2 with Fable 5.1, Claude Opus 5.5, GPT-6 Astra and Gemini 4 Carbon. Scores use each model’s highest available effort unless noted, with safeguards on for the Claude models.

From Anthropic’s launch post. On Terminal-Bench-Science, Mythos 5.2 scores 75.3%. Dashes: not reported.

3 more rows cover Mythos 5.2 only, because Fable 5.2’s safeguards block most cybersecurity and research-biology tasks: CyScenarioBench 76.3% (Mythos 5.1: 61.7%; Opus 5.5 without safeguards: 67.6%; GPT-6 Cyber: 61%), CVE Rediscovery 89.6% (78.3%; 81.3%) and 24-hour de novo protein binder design 85.1% (79.1%; 82.6%).

The biggest jumps over Fable 5.1 are agentic: Remote Labor Index from 17.9% to 41.3%, RoboDojo from 27.6% to 64.9%, FrontierCode from 32.4% to 49.1%. Anthropic also says that at high effort Fable 5.2 scores 45.3% on FrontierCode, level with Opus 5.5’s best 45.5%, for roughly half the cost per task.

The 2 rows it loses

AutomationBench goes to Gemini 4 Carbon, 53.9% to 51.4%. That’s the business-workflow row, and the only one where a non-Anthropic model beats Fable 5.2. Fable 5.2 still gains 20 points over Fable 5.1’s 31.4% there.

AA-Briefcase Elo goes to Opus 5.5, 1800 to 1785. On the same benchmark’s rubric score, Fable 5.2 leads 66.1% to 61.2%. The rubric grades each answer against a checklist; an Elo ranks models by pairwise comparisons. Opus 5.5, at half the price per token, wins slightly more of those matchups.

Independent numbers: #1 on both boards

Two independent evaluators posted results on launch day. Both put Fable 5.2 first overall.

Artificial Analysis leaderboard and model pages; Vals.ai model page and leaderboards updated November 9, 2026. Vals.ai ran Fable 5.2 at high effort.

For vibe coders, Vibe Code Bench is the closest match: the model builds a working web app from a plain-language spec, and the app is checked by point-and-click testing. Fable 5.2 ranks #1 of 127 and costs 34% less per test than Fable 5.1 did. On Terminal-Bench 4.0, though, it only ties Opus 5.5 and costs $2.52 more per test.

Why a cheaper model costs 2x per task

The price per token fell. The number of tokens rose faster. To run the Intelligence Index at max effort, Fable 5.2 produced 275M output tokens to Fable 5.1’s 163M, and processed 28.3B input tokens to 9.46B. Artificial Analysis also measured 44.4 minutes per task to Fable 5.1’s 22.5.

The fix is the effort dial. Artificial Analysis ran Fable 5.2 at 5 settings:

Artificial Analysis leaderboard, Intelligence Index v5.0.1.

This is the second Fable release in a row where the headline price and the measured bill point in opposite directions. We covered the first one, a 75% cache discount on Fable 5.1, in Anthropic Cut a Price 75%. Its Own Benchmark Partner Says You Pay 20% More.

Subscriptions: a free week for Pro, resets for Max and Ultra

Fable 5.2 is in the API and on AWS, Google Cloud and Microsoft Azure from day one. Claude subscribers get a launch offer, per Anthropic’s help center:

A reset gives back the weekly usage you spent on Fable models and refills the 5-hour window, if your plan has one. It doesn’t raise your limits. Only accounts with paid billing in place before November 9 at 10:00 am PT qualify. After the week, Fable on Pro goes back to usage credits.

What this means if you build with AI

Here is what the price cut does to one agent session billed per token, through the API or a tool like Cursor with your own key. Our example: 2M input tokens, 1.8M of them cache hits, and 100K output tokens. We leave out cache-write premiums to keep the math readable.

Our math from list prices. The 1.69x row applies the ratio Artificial Analysis measured at max effort (275M ÷ 163M output tokens).

Token for token, you save 20%: $1.49 on this session. If Fable 5.2 at max writes 1.69x the output, the same session costs $8.71, more than Fable 5.1 did. The effort setting decides whether the cut reaches your bill.

Default to Opus 5.5. Anthropic says so itself: start most coding and knowledge work on Opus 5.5 and move up to Fable when the task demands it. It ties Fable 5.2 on Terminal-Bench 4.0, wins SWE-bench Multimodal and costs half as much per token.

Use Fable 5.2 at high for the hard part. Long migrations, ambiguous specs, the bug the cheaper model couldn’t fix. At high it beat Opus 5.5 at max on the Intelligence Index for less money. Reach for xhigh or max only when a wrong answer costs more than $26.

Use GPT-6.1 Sol for volume. At $2 / $10 it scores 55 on the index for $1.27 a task, and 88.93% on Vibe Code Bench for $6.23 a test. For bulk edits, tests and refactors, that’s within 10 index points of Fable 5.2 at max for under 5% of its cost per task.

Fable 5.2 is available on MegaBrain as anthropic/claude-fable-5.2, next to Opus 5.5, GPT-6.1 Sol and the other text models: one OpenAI-compatible API, one key and one balance, which also covers FLUX, Nano Banana, Seedream and Kling for images and video. Switching between the three models above is a change to the model string.

Sources

Pages opened November 10–11, 2026. Every number above comes from these pages, or from our arithmetic on them where we say so.

• Anthropic — Introducing Claude Fable 5.2 and Claude Mythos 5.2 (November 9, 2026: price, speed, benchmark table, effort notes, availability)
• Anthropic — Claude Fable 5.2 and Claude Mythos 5.2 System Card
• Claude API docs — Pricing (Fable 5.2, Fable 5.1, Opus 5.5: base, cache, batch and fast mode rates)
• Claude API docs — Models overview (model ID, context window, max output, default effort)
• Claude Help Center — Claude Fable 5.2 launch offers (resets and the Pro week)
• Artificial Analysis — LLM leaderboard (Intelligence Index and cost per task by effort)
• Artificial Analysis — Claude Fable 5.2 (max) model page
• Artificial Analysis — Claude Fable 5.1 (max) model page
• Vals.ai — Claude Fable 5.2 model page (Vals Index and component ranks)
• Vals.ai — Terminal-Bench 4.0 leaderboard (updated November 9, 2026)
• Vals.ai — Vibe Code Bench v1.1 leaderboard (updated November 9, 2026)
• OpenAI — GPT-6.1 Sol model docs (pricing and long-context rates)

Full disclosure: MegaBrain sells pay-as-you-go API access to image, video and text models, including Claude Fable 5.2. MegaBrain is not affiliated with Anthropic, OpenAI, Google, Artificial Analysis or Vals.ai.