MLS Bench Lite — Kimi K2.7 Code source; own Lite benchmark, not the MLS (LibriSpeech) ASR corpus despite acronym.
| # | Model | Vendor | Best score | Runs | Last seen |
|---|---|---|---|---|---|
| 1 | Claude Fable 5 | Anthropic | 49.9 | 2 | 2026-08-12 |
| 2 | Kimi K3 | Moonshot | 48.3 | 2 | 2026-08-23 |
| 3 | GPT-5.6 Sol | OpenAI | 46.2 | 2 | 2026-08-12 |
| 4 | Claude Opus 4.8 | Anthropic | 42.8 | 3 | 2026-08-12 |
| 5 | Qwen3.8 Max Preview | Alibaba | 41 | 2 | 2026-08-23 |
| 6 | GLM-5.2 Full Open Source | Z.ai | 40.4 | 1 | 2026-07-27 |
| 7 | GPT-5.5 | OpenAI | 35.5 | 2 | 2026-07-27 |
| 8 | Kimi K2.7 Code | Moonshot | 35.1 | 2 | 2026-08-23 |
| 9 | Qwen3.7 Max | Alibaba | 31.7 | 1 | 2026-08-12 |
Best tracked score per model (default configuration; source-attributed and verification-tiered). Open a model for its full benchmark surface, provenance and pricing.