GSM8K
Harbor adapterGSM8K: 小学应用题（多步推理）
- Samples
- 1,319
- Released
- 2021
- Scores
- 180
How to run
ltb · theoremqa
$ npm i -g lettokenburn-cli$ ltb bench pull theoremqa --n 50$ ltb bench run ./theoremqa -a claude-code -m anthropic/claude-opus-4-8$ ltb bench push
Runs happen on your machine with your own subscription; only the score summary is pushed. CLI docs
Scores
8 sourced entries| Harness | Source | Confidence | Link | |||
|---|---|---|---|---|---|---|
| GPT-4 | 43.8 | N/A | Independent | medium | May 21, 2023 | arxiv.org |
| PaLM-2-unicorn | 31.8 | N/A | Independent | medium | May 21, 2023 | arxiv.org |
| GPT-3.5 Turbo | 30.2 | N/A | Independent | medium | May 21, 2023 | arxiv.org |
| Claude-v1 | 24.9 | N/A | Independent | medium | May 21, 2023 | arxiv.org |
| code-davinci-002 | 23.9 | N/A | Independent | medium | May 21, 2023 | arxiv.org |
| Claude-instant | 23.6 | N/A | Independent | medium | May 21, 2023 | arxiv.org |
| text-davinci-003 | 22.8 | N/A | Independent | medium | May 21, 2023 | arxiv.org |
| PaLM-2-bison | 21.0 | N/A | Independent | medium | May 21, 2023 | arxiv.org |
Every number links to where it was published. Low-confidence rows are muted and need per-cell verification.
Reproduced by the community
All runsLoading community runs…
About
TheoremQA (Theorem QA) — 定理应用推理. Scoring: 准确率. Difficulty: 高. Imported from benchmark_research.md (2026-06).