Claude Fable 5 Launch: Double‑Price, Explosive Performance Gains

Claude Fable 5 was officially released, continuing Anthropic’s tradition of naming models after literary concepts ("Fable" meaning a story or allegory).

Pricing

The model costs $10 per million input tokens and $50 per million output tokens, which is roughly double the price of the previous Opus 4.8 model.

Benchmark Performance

On the SWE‑bench Pro programming evaluation, Fable 5 achieved an 80.3% success rate, compared with 69.2% for Opus 4.8 and 58.6% for GPT‑5.5. On the stricter SWE‑bench Verified suite, it reached 95.0%, the highest score observed to date.

Karpathy, testing the model immediately after release, described the improvement as a "leap worthy of a major version upgrade," comparable to the jump seen with Claude 4.5.

Real‑World Case Study

Stripe migrated a 50‑million‑line Ruby codebase. The internal engineering team required over two months to complete the migration manually, whereas Claude Fable 5 finished the task in a single day.

Frontend Evaluation

In the ViBench frontend benchmark, Fable 5 nearly filled all basic frontend use‑case slots, effectively achieving the "one‑line‑to‑an‑app" capability.

Cost Considerations

Although the model’s raw performance is superior, some medium‑size programming or data‑analysis tasks silently consume 0.5–1 million tokens, making the per‑task cost several times higher than Opus 4.8.

Free usage was available until June 22; starting June 23, users must pay usage credits, limiting the window for cost‑free experimentation.

Upgrade Recommendation

For complex, long‑chain tasks, large refactorings, or workloads that run for many hours, Fable 5’s advantages are tangible and the higher cost can be justified. Simple Q&A or routine small requests are adequately handled by Opus 4.8, making the upgrade unnecessary for those scenarios.