Best LLM for Constrained Rewriting (2026)
After testing 5 models on real constrained rewriting tasks, GPT-5 Nano leads with a score of 3/3 at $0.40/MTok output.
Which Models Are Best for Constrained Rewriting?
| # | Model | Score | Output $/MTok | Bracket |
|---|---|---|---|---|
| 1 | GPT-5 Nano | 3/3 | $0.40 | Budget |
| 2 | Mistral Small 4 | 3/3 | $0.60 | Budget |
| 3 | Ministral 3 14B | 2/3 | $0.20 | Budget |
| 4 | Mistral Small 3.2 | 2/3 | $0.20 | Budget |
| 5 | Grok 4.1 Fast | 2/3 | $0.50 | Value |
How Do We Test for Constrained Rewriting?
Compression within hard character limits
Test name: constrained_rewriting. Scored 1-3 by LLM-as-judge. Full methodology →