DEV Community

Shivam Kumar
Shivam Kumar

Posted on

Fable 5.2 Is 30% Faster and 20% Cheaper — Here's the Migration Math

Fable 5.2 Is 30% Faster and 20% Cheaper — Here's the Migration Math

Shivam Kumar (VisionQuantech) — 17 November 2026

Anthropic shipped Claude Fable 5.2 and Mythos 5.2 on 9 November. Same weights, two safeguard profiles: Fable is generally available, Mythos goes only to organisations vetted through Anthropic's Cyber and Life Sciences Verification Programs. The API ID is claude-fable-5-2, on the Claude Platform, Bedrock, Google Cloud, and Microsoft Foundry.

Here is what changes for your code, your bill, and your evals.

Price and speed

  • $8/$40 per million tokens (input/output), down from $10/$50 — a 20% cut.
  • Up to 30% faster outputs than Fable 5.1.
  • Fast mode: $16/$80, up to 2.5x token generation speed.
  • Specs unchanged: 1M context, 128K max output, adaptive thinking always on, default effort High, June 2026 knowledge cutoff.

Benchmarks (Anthropic-reported)

FrontierCode 2.0: 49.1% at Xhigh effort (Fable 5.1: 32.4%, Opus 5.5: 45.5%, GPT-6 Astra: 42.8%). FrontierSWE v2: 69.9% (5.1: 56.3%). RoboDojo: 64.9% (5.1: 27.6%). Humanity's Last Exam with tools: 74.2%. OSWorld 2.1 offline: 89.3%. Epoch Capabilities Index: a record 174.

Two counterpoints from Anthropic's own table: on AutomationBench, Gemini 4 Carbon (53.9%) beats Fable 5.2 (51.4%); on the AA-Briefcase knowledge-work Elo, Opus 5.5 (1800) edges Fable 5.2 (1785). Anthropic still recommends Opus 5.5 as the daily driver.

The migration math

Take a realistic agent workload: 10M input + 2M output tokens per day.

  • Fable 5.1: 10 × $10 + 2 × $50 = $200/day → $6,000/month
  • Fable 5.2: 10 × $8 + 2 × $40 = $160/day → $4,800/month
  • Saving: $1,200/month, $14,400/year — on a better model.

But the real saving is cost per task, not per token. At High effort, Fable 5.2 matches Opus 5.5's best FrontierCode score for roughly half the cost per task. On Humanity's Last Exam and Chartography, lower effort tiers match Opus's best for under a tenth of the cost. Test Medium effort on Fable 5.2 against your current High-effort Opus setup before you conclude anything about your bill.

For context, Qwen4-Max — which claimed the coding crown over Fable 5.1 last week — lists at $1.50/$4.50, so the same workload is $24/day. Fable 5.2 is still ~6.7x pricier per token, and its crown changed hands twice in eight days. Price per token is the wrong metric; price per completed task is the one that pays your invoice.

The deliberate cap

Anthropic limited Fable 5.2's skill at accelerator kernel optimisation and LLM architecture work — the tasks that could most directly speed up AI development itself. There it performs at about Opus 5.5's level, by design. If your workload is kernel tuning or architecture research, migrating buys you nothing. For everything else, it is the stronger model at a lower price.

Migration checklist

  1. Swap the model ID to claude-fable-5-2 and re-run your evals — adaptive thinking cannot be disabled and default High effort changes token usage.
  2. Re-run any Qwen4-Max vs Fable 5.1 comparison from last week; it is already stale.
  3. Use fast mode ($16/$80) only where a human waits on the model — interactive coding, live demos. Batch jobs stay on standard.
  4. Test Medium effort first. That is where the hidden saving lives.

Fable 5.2 is the strongest model Anthropic has shipped — cheaper, faster, and honest about its own limit. Migrate with evals, not with enthusiasm.


Sources: Anthropic launch announcement and system card (9 Nov 2026); Anthropic model docs; Epoch AI. Benchmarks are vendor-reported.

Top comments (0)