Claude Fable 5.5 (Leaked Checkpoint) vs TypeSafe AI Jev
A direct, empirical evaluation of Claude Fable 5.5 (Leaked Checkpoint) (Anthropic) against TypeSafe AI Jev (TypeSafe AI (Diogo Almeida)).
Claude Fable 5.5 (Leaked Checkpoint)
Developed by Anthropic · Proprietary Frontier Dynamic Reasoner parameters
Terminal-Bench 4.0 autonomous shell execution (78.5% target), recursive self-critique, and 1.5M-token repository refactoring.
TypeSafe AI Jev
Developed by TypeSafe AI (Diogo Almeida) · Non-Autoregressive System 1 Transformer parameters
Sub-15ms operational tasks, 100% schema-valid JSON extraction, deterministic tool dispatch, and high-throughput agent routing.
| Metric | Claude Fable 5.5 (Leaked Checkpoint) | TypeSafe AI Jev |
|---|---|---|
| SWE-bench Verified | 84.2% | 68.4% |
| MMLU-Pro | 94.6% | 84.6% |
| MATH-500 | 98.4% | 91.2% |
| Context Window | 1,500,000 tokens | 32,000 tokens |
| Input Pricing (per 1M) | $3.50 | $0.06 |
| Output Pricing (per 1M) | $17.50 | $0.12 |