Benchmarks › IFBench
IFBench
Follow precise formatting, counting and text instructions
83.3top score · Grok 4.3
245models ranked
178open-weights models
Prompt-level loose accuracyhigher is better
Oct 20, 2026last updated
How the test works
1
Task
294 tasks
- What the model doesFollow precise formatting, counting and text instructions
- Kinds of tasksSingle-turn instruction following
- ExampleIs it plausible that frequent hardship can make a society more resilient? Include exactly 2 numbers in the response.
2
Run
5 runs per task
- Token budgetNon-reasoning: up to 16,384 output tokens, adjusted to model limits; reasoning: model-specific maximum
- Temperature0 for non-reasoning; 0.6 for reasoning, unless the model lab recommends otherwise
3
Score
Prompt-level loose accuracy
- Graded byOfficial allenai/IFBench rule verifiers in loose mode
- Better isHigher
Sources: artificialanalysis.ai, github.com
IFBench leaderboardtop 10 of 245
Full leaderboard with filters →| Model | IFBench | Per task | Price | Speed | Supports | Context |
|---|---|---|---|---|---|---|
83.3medium | — | $1.25 / $2.50 | 93 | 1M | ||
82.9thinking | $0.51 | $0.30 / $1.20 | 157 | 524K | ||
81.4thinking | $0.58 | $0.50 / $2.50 | 138 | 1M | ||
80.5thinking | $1.15 | $1.48 / $4.43 | 201 | 1M | ||
80.4 | — | — | 1M | |||
79.9thinking | $0.054 | $0.43 / $0.87 | 46 | 1M | ||
79.6low | — | 134 | 256K | |||
78.8thinking | $0.47 | $0.60 / $3.60 | 55 | 262K | ||
78thinking | — | $0.50 / $3 | 211 | 1M | ||
78thinking | $0.22 | $0.32 / $1.28 | 58 | 1M |
IFBench questions
What is IFBench?›
Follow precise formatting, counting and text instructions. It has 294 tasks. Created by Allen Institute for AI.
Which model is best on IFBench?›
Grok 4.3 leads IFBench with 83.3, ahead of MiniMax M3 (82.9) and Nemotron 3 Ultra (81.4).
What is the best open-weights model on IFBench?›
MiniMax M3 with 82.9, then Nemotron 3 Ultra (81.4).
How is IFBench scored?›
Prompt-level loose accuracy; official allenai/ifbench rule verifiers in loose mode; 5 runs per task per task.