Rank 01 / 10Baseline
anthropic/claude-sonnet-latest
90%Pass · 9/10 runs
- Avg cost / run
- $0.028
- P50 latency
- 8.8s
- Cost / pass
- $0.031
The control group. Expensive, boring, correct.
SPEC-01Full report →
Please wait while we prepare your content.
One card per model, scored against our real work across five task types. Baselines are marked — they set the bar, they don't win by default.
The control group. Expensive, boring, correct.
Best non-baseline on code. Would use again.
Steady across every task. Slows on long context.
Fast baseline. Weak on agentic runs.
Slow but stubborn. Finishes agentic work others abandon.
Best cost-per-pass on the bench.
Cheap. Drops details under pressure.
Great prose, unreliable structure.
Cold-start latency ruins it for interactive use.
Fast and cheap. Fails anything subtle.