benchmark evidence
TutorBench
Scale Labs tutoring benchmark measuring adaptive explanations, feedback, and hint quality.
activeupstream source ->
winner on TutorBench
OpenAI: GPT-577.3
direct benchmark result, not a broad vertical composite | source row dated 2025-09-10
scored on 2025-12-15 · stale source data (165d)
latest mapped results | top 20
| # | Model | Score | Evidence | Tested |
|---|---|---|---|---|
| 1 | OpenAI: GPT-5 | 77.3 | model-only independent_benchmark | 2025-09-10 |
| 2 | OpenAI: GPT-5.2 | 74.7 | model-only independent_benchmark | 2025-12-15 |
| 3 | Anthropic: Claude Opus 4.5 | 69.6 | model-only independent_benchmark | 2025-09-10 |
| 4 | Anthropic: Claude Sonnet 4.5 | 63.8 | model-only independent_benchmark | 2025-09-10 |
| 5 | Meta: Llama 4 Maverick | 56.2 | model-only independent_benchmark | 2025-09-10 |
what this result means
Scale Labs tutoring benchmark measuring adaptive explanations, feedback, and hint quality.
This benchmark contributes direct public evidence. Read its scope before generalizing the result.
A win here is a win on TutorBench. Broad task pages require independent corroboration before naming a general winner.
source record
category: writing
metric: accuracy
matched models: 5
latest source date: 2025-12-15
direction: higher is better