Claude Fable 5.2: A Lower Token Price, a Higher Bill Per Task

Claude Fable 5.2 lists at $8 per million input tokens and $40 per million output tokens, down from $10 and $50 for its predecessors. Yet Artificial Analysis measured its peak cost per task at $26.50, roughly double Fable 5.1’s $12.99. For engineering leads budgeting agentic AI workloads, both numbers are true, and only one of them shows up on the invoice.

What Happened

Anthropic announced Claude Fable 5.2 and Claude Mythos 5.2 on November 9, describing them as “the same underlying model.” Fable 5.2 is available on all platforms, including Amazon Web Services, Google Cloud and Microsoft Azure, under the API ID claude-fable-5-2. Mythos 5.2 is limited to organizations vetted through Anthropic’s Cyber Verification Program or Life Sciences Verification Program.

Anthropic says output is up to 30% faster than Fable 5.1, and an optional fast mode runs up to 2.5 times faster at $16 input and $80 output per million tokens. Its own benchmark table puts Fable 5.2 at 49.1% on FrontierCode 2.0, against 32.4% for Fable 5.1 and 45.5% for Claude Opus 5.5. On FrontierSWE v2 the scores are 69.9%, 56.3% and 62.3%. On AutomationBench they are 51.4%, 31.4% and 42.5%. Anthropic also says the safeguards now produce about 55% fewer blocks than Fable 5.1’s.

Why It Matters

The Decoder’s analysis draws on Artificial Analysis, whose Intelligence Index v5 ranks Fable 5.2 first with 65 points at its highest compute tier. Gemini 4 Carbon scores 60 and Opus 5.5 scores 59. The cost of that lead is the story. At peak effort, Fable 5.2 averaged $26.50 per task, about 2.6 times Opus 5.5’s $10. That holds even though Fable 5.2 used 98,000 output tokens per task, against 173,000 for Opus 5.5 at its Max tier.

So the per-token cut doesn’t translate into a cheaper workload when a team runs the model at its highest setting. What drives the bill is the effort tier a task runs at. Anthropic’s own guidance points the same way. It reports that at High effort Fable 5.2 scores 45.3% on FrontierCode, against Opus 5.5’s 45.5%, “for roughly half the cost per task.” It still calls Opus 5.5 its “strongest daily driver” for most work.

The lead isn’t uniform either. The Decoder notes Fable 5.2 trails on agentic knowledge work, at 1,785 Elo on AA-Briefcase against 1,800 for Opus 5.5. On Terminal-Bench 4.0 it scores 60%, tied with Opus 5.5 and behind Kimi K3.1 Pro at 67%.

What This Means for Builders

The pattern from Fable 5’s launch repeats: the top score comes at a premium, and the premium depends on configuration. Teams routing agent work should treat effort tier as a cost control, not a quality slider to leave at maximum. Hard tasks, such as long coding runs or formal proofs, may justify peak effort. The Decoder reports that about 1,000 Fable 5.2 agents formalized Fermat’s Last Theorem in Lean in 30 hours. Routine tool calls and summaries rarely need it.

Source: The Decoder

Content created with AI assistance and reviewed for accuracy.