Deep reasoning
Tackle novel, open-ended problems.
Frontier AI research for deeper, more general reasoning.
A frontier model suite for complex problems with uncertain or incomplete information.
Tackle novel, open-ended problems.
Produce rigorous, checkable results.
Follow intent with robust safeguards.
A reasoning foundation for scientific discovery across complex, long-horizon problems.
Read sustained theorem accuracy across hundreds of verified derivation steps.
Explore proof benchmarks
| Capability | Abyss-1 Pro | Abyss-1 | Abyss-1 Fast |
|---|---|---|---|
| Context window | 256k | 256k | 128k |
| MATH-500 | 94.1% | 92.7% | 88.4% |
| GPQA Diamond | 90.2% | 88.4% | 82.1% |
| MiniF2F Lean 4 | 95.1% | 91.8% | 84.6% |
| Throughput | 42 tok/s | 64 tok/s | 118 tok/s |
| Formal proof verifier | — | ||
| Private weight deploy |
Frontier AI research for deeper, more general reasoning.