benchmark evidence
BenchLM Instruction Following
BenchLM instruction following score (0-100).
activeupstream source ->
winner on BenchLM Instruction Following
OpenAI: GPT-5.492.6
direct benchmark result, not a broad vertical composite | source row dated 2026-07-09
scored on 2026-07-09
latest mapped results | top 20
| # | Model | Score | Evidence | Tested |
|---|---|---|---|---|
| 1 | OpenAI: GPT-5.4 | 92.6 | model-only aggregator | 2026-07-09 |
| 2 | Anthropic: Claude Opus 4.6 | 91.3 | model-only aggregator | 2026-07-09 |
| 3 | Anthropic: Claude Sonnet 4.6 | 79.9 | model-only aggregator | 2026-07-09 |
| 4 | Anthropic: Claude Sonnet 4.5 | 79.5 | model-only aggregator | 2026-07-09 |
| 5 | OpenAI: GPT-5.2 | 79.1 | model-only aggregator | 2026-07-09 |
| 6 | Z.ai: GLM 5 | 76.0 | model-only aggregator | 2026-07-09 |
| 7 | OpenAI: GPT-4.1 | 73.2 | model-only aggregator | 2026-07-09 |
| 8 | OpenAI: GPT-5.1 | 72.7 | model-only aggregator | 2026-07-09 |
| 9 | OpenAI: GPT-5.4 Mini | 70.7 | model-only aggregator | 2026-07-09 |
| 10 | Google: Gemini 2.5 Pro | 62.1 | model-only aggregator | 2026-07-09 |
| 11 | Anthropic: Claude Opus 4.5 | 42.4 | model-only aggregator | 2026-07-09 |
what this result means
BenchLM instruction following score (0-100).
This benchmark contributes direct public evidence. Read its scope before generalizing the result.
A win here is a win on BenchLM Instruction Following. Broad task pages require independent corroboration before naming a general winner.
source record
category: writing
metric: score
matched models: 11
latest source date: 2026-07-09
direction: higher is better