LetTokenBurnEHR-Bench: A Chinese Multi-Task Benchmark for Recruitment AI
An open Chinese benchmark with 8,336 instances across 12 recruitment tasks (generation, extraction, classification, retrieval, ranking, similarity) evaluating LLMs on enterprise hiring scenarios.
$ npm i -g lettokenburn-cli$ ltb bench info ehr-bench
External harness. No Harbor adapter has been generated for this benchmark yet. `ltb bench info` prints the known links.
Runs happen on your machine with your own subscription; only the score summary is pushed. CLI docs
| Harness | Source | Confidence | Link | |||
|---|---|---|---|---|---|---|
| GPT-5.5 | 95.2 | N/A | Independent | medium | Oct 12, 2026 | arxiv.org |
| Gemini-3.5-flash | 94.4 | N/A | Independent | medium | Oct 12, 2026 | arxiv.org |
| Kimi 2.6 | 93.8 | N/A | Independent | medium | Oct 12, 2026 | arxiv.org |
| Qwen3.8 Max | 93.8 | N/A | Independent | medium | Oct 12, 2026 | arxiv.org |
| Claude Opus 4.8 | 93.5 | N/A | Independent | medium | Oct 12, 2026 | arxiv.org |
| GLM-5.2 | 93.2 | N/A | Independent | medium | Oct 12, 2026 | arxiv.org |
| GLM-5.3 | 92.8 | N/A | Independent | medium | Oct 12, 2026 | arxiv.org |
| DeepSeek V4 Pro | 86.0 | N/A | Independent | medium | Oct 12, 2026 | arxiv.org |
| MiniMax-M3 | 83.6 | N/A | Independent | medium | Oct 12, 2026 | arxiv.org |
| Kimi K3 | 65.7 | N/A | Independent | medium | Oct 12, 2026 | arxiv.org |
| Claude Opus 5 | 64.3 | N/A | Independent | medium | Oct 12, 2026 | arxiv.org |
| GPT-5.6 Sol | 43.8 | N/A | Independent | medium | Oct 12, 2026 | arxiv.org |
Every number links to where it was published. Low-confidence rows are muted and need per-cell verification.
EHR-Bench comprises 8,336 task instances derived from public materials and controlled augmentation, guided by an internal recruitment dataset. It covers twelve tasks across six capability families including generation, extraction, and ranking.
Scoring uses task-specific metrics like ROUGE-L, rubric scores, and accuracy with source-aware expert review. Evaluation of twelve LLMs reveals distinct task strengths and consistent overall ordering against private operational references.