Model Release Radar
Price/capability frontier position, benchmark scores, and community signal
Model Release Radar
Kimi K3
moonshot·open weights·Kimi K3 license·released Thursday, Jul 16, 2026
Frontier position
- Behind frontier DeepSWE pass@1 (agentic coding) - GLM-5.3, GPT-5.6 Terra, GPT-5.6 Sol, Claude Opus 5 are priced the same or lower and at least as capable.
Cost vs. capability
Loading chart…
Benchmark scores
Measured cost per task (DeepSWE)
| Benchmark | Pass@1 | Measured cost / task | Frontier |
|---|---|---|---|
| DeepSWE (agentic coding) (max effort, 95% CI 64.0%-73.0%, 4 runs) | 68.5% | $4.65 (median $3.60) | behind |
Artificial Analysis scores (no matched cost)
Artificial Analysis scores below have no measured per-task cost, so no frontier claim is made for them here.
| Benchmark | Score |
|---|---|
| AA intelligence index | 59.7 |
| AA coding index | 76.2 |
| GPQA Diamond (science) | 93.5% |
| Humanity's Last Exam | 46.9% |
| LCR | 82.7% |
| SciCode (coding) | 58.7% |
| Tau2 (banking) | 46.0% |
| Terminal-Bench 2.1 (agentic) | 85.0% |
Variants
This page covers 2 reasoning-effort variants of the same model.
| Variant | AA coding index | AA intelligence index | Price /1M |
|---|---|---|---|
| Max effort (shown above) | 76.2 | 59.7 | $6.00/1M |
| Low effort | 72.0 | 48.3 | $6.00/1M |
Community signal
| Variant | Coding Elo | Overall Elo | Votes |
|---|---|---|---|
| Max effort | 1,512 | 1,476 | 16,586 |