The Pile (test split, bits-per-byte) — Pile test-set LM-loss metric (BPB), not accuracy-scale.
| # | Model | Vendor | Best score | Runs | Last seen |
|---|---|---|---|---|---|
| 1 | Qwen2.5 Instruct 72B | Alibaba | 0.6 | 1 | 2026-05-01 |
| 2 | DeepSeek-V2 | DeepSeek | 0.6 | 1 | 2026-05-01 |
| 3 | DeepSeek-V3 | DeepSeek | 0.5 | 1 | 2026-05-01 |
| 4 | Llama 3.1 Instruct 405B | Meta | 0.5 | 1 | 2026-05-01 |
Best tracked score per model (default configuration; source-attributed and verification-tiered). Open a model for its full benchmark surface, provenance and pricing.