Skip to content

Head-to-head · 2026

Claude Fable 5 vs Claude Opus 4.8

Verdict. Fable 5 takes agentic coding, constraint precision, and hard reasoning — at double the price. Opus 4.8 keeps situational judgment, code-craft quality, and the better $/token for human-read prose. Route by task shape, not by leaderboard.

Claude Fable 5Claude Opus 4.8
ProviderAnthropicAnthropic
Released2026-06-092026-05-28
Context1M1M
Max output128K128K
Input /1M$10.00$5.00
Output /1M$50.00$25.00
Modalitiestext, vision, codetext, vision, code

The analysis

This is the rare comparison with first-party data behind it: we ran four head-to-head eval rounds inside Claude Code within 24 hours of the Fable 5 launch, with published JSON receipts. The split is consistent — Fable 5 was the most constraint-compliant model in every round (7/7 on a script-verified constraint stack Opus failed) and the only one to solve a hard no-tools reasoning task Opus got confidently wrong. Opus 4.8 flagged a governance-gated edit Fable executed silently, pushed back on a contradictory spec, and won the real-world component build on accessibility craft.

On vendor numbers, Fable 5 is the new agentic ceiling: 95.0% SWE-Bench Verified and ~80% SWE-Bench Pro at launch (vendor-claimed) against Opus 4.8’s well-corroborated 88.6% / 69.2%. Anthropic positions the gap as widening with task length — which matches the agentic shape of the launch card.

Pricing is the routing fork: Fable 5 at $10/$50 per million tokens is exactly double Opus 4.8’s unchanged $5/$25 (and equal to Opus fast mode). The 2× pays for itself in agentic pipelines where constraint precision and task length bind; it does not pay for itself in judgment-heavy review or prose — our blind style verdicts flipped between rounds.

Pick Claude Fable 5 if…

  • Agentic pipelines feeding schemas, tools, and other agents
  • Long-horizon, multi-file coding (the SWE-Bench Pro lead is this shape)
  • Strict output contracts where a leaked preamble is a failed tool call

Pick Claude Opus 4.8 if…

  • Ambiguous or possibly-wrong specs — it pushes back instead of executing
  • Code review and accessibility-critical component work
  • Human-read prose and deep single-shot analysis at half the price

Claude Fable 5

Mythos-class made generally available — the new agentic-coding ceiling, at 2× Opus pricing.

Claude Opus 4.8

Modest version bump, real frontier gains — tops the intelligence index at the same price as 4.7.

More comparisons