IFEval (HELM)
As of 29 Oct 2026, the top score on IFEval (HELM) is 95.1% by Grok 3 mini (SpaceXAI), ahead of Grok 4 at 94.9%, across 68 model configurations with a published result (HELM Capabilities, result dated 24 Nov 2025). IFEval gives the model writing tasks with verifiable instructions attached: respond in exactly three paragraphs, avoid the letter e, end with a given phrase, and so on. Each instruction can be checked by a program, so no judge model is needed. IFEval (HELM) counts toward the BenchLeader instruction following score.
Key facts (as of 29 Oct 2026)
- Grok 3 mini holds the top IFEval (HELM) score at 95.1% (HELM Capabilities, 24 Nov 2025).
- Grok 4 is second on IFEval (HELM) at 94.9% (HELM Capabilities, 24 Nov 2025).
- IFEval (HELM): Strict instruction following on verifiable constraints. Run by Stanford CRFM's HELM (HELM Capabilities).
- 68 model configurations have a published IFEval (HELM) result on BenchLeader (BenchLeader, 29 Oct 2026).
- IFEval (HELM) feeds the instruction following category of the BenchLeader Index (BenchLeader methodology).
Strict instruction following on verifiable constraints. Run by Stanford CRFM's HELM.
- Published by
- HELM Capabilities
- Category
- Instruction following
- Index weight
- 1.0
- Models
- 68
- Data as of
- 29 Oct 2026
- Newest result
- 24 Nov 2025
The publisher has posted no new results on this board since 24 Nov 2025. BenchLeader re-reads it every day, so newer models appear here as soon as they are published.
Stanford Center for Research on Foundation Models (crfm.stanford.edu), Apache-2.0 results.
What the test looks like
IFEval gives the model writing tasks with verifiable instructions attached: respond in exactly three paragraphs, avoid the letter e, end with a given phrase, and so on. Each instruction can be checked by a program, so no judge model is needed.
How it is scored
Strict accuracy: the share of instructions satisfied exactly, as run by Stanford CRFM's HELM in its fixed harness.
What to keep in mind
It checks whether rules were obeyed, not whether the writing was any good, and the constraints are artificial. Near saturation for frontier models, so it mainly separates the mid-field.
- 1Grok 3 mini95.1%
- 2Grok 494.9%
- 3GPT-5.193.5%
- 4GPT-5 nano93.2%
- 5o4-mini92.9%
- 6GPT-5 mini92.7%
- 7Claude Opus 491.8%
- 8Llama 4 Maverick90.8%
- 9GPT-4.1 mini90.4%
- 10Gemini 2.5 Flash89.8%
- 11Granite 4.0 Small89.0%
- 12Grok 388.4%
- 13Gemini 3 Pro87.6%
- 14Mistral Large 2.187.6%
- 15GPT-587.5%
| # | ||||
|---|---|---|---|---|
| 1 |  Grok 3 mini | 95.1% | 52.3 | 2025-11-24 |
| 2 | Grok 4 | 94.9% | 58.4 | 2025-11-24 |
| 3 |  GPT-5.1 | 93.5% | 56.8 | 2025-11-24 |
| 4 | GPT-5 nano | 93.2% | 55.9 | 2025-11-24 |
| 5 | o4-mini | 92.9% | 56.9 | 2025-11-24 |
| 6 | GPT-5 mini | 92.7% | 56.4 | 2025-11-24 |
| 7 |  Claude Opus 4 | 91.8% | 52.6 | 2025-11-24 |
| 8 |  Llama 4 Maverickopen ↗ | 90.8% | 42.2 | 2025-11-24 |
| 9 | GPT-4.1 mini | 90.4% | 44.8 | 2025-11-24 |
| 10 |  Gemini 2.5 Flash | 89.8% | 52.7 | 2025-11-24 |
| 11 |  Granite 4.0 Small | 89.0% | – | 2025-11-24 |
| 12 | Grok 3 | 88.4% | 52.5 | 2025-11-24 |
| 13 | Gemini 3 Pro | 87.6% | 60.2 | 2025-11-24 |
| 14 |  Mistral Large 2.1open | 87.6% | 42.6 | 2025-11-24 |
| 15 | GPT-5 | 87.5% | 57.7 | 2025-11-24 |
| 16 |  Palmyra X 004 | 87.2% | – | 2025-11-24 |
| 17 | o3 | 86.9% | 61.0 | 2025-11-24 |
| 18 | Claude 3.5 Sonnet | 85.6% | 47.1 | 2025-11-24 |
| 19 | Claude Sonnet 4.5 | 85.0% | 53.7 | 2025-11-24 |
| 20 |  Kimi K2open ↗ | 85.0% | 51.0 | 2025-11-24 |
| 21 | Claude Opus 4thinking | 84.9% | 55.2 | 2025-11-24 |
| 22 | granite-4.0-microopen ↗ | 84.9% | 36.4 | 2025-11-24 |
| 23 | GPT-4.1 nano | 84.3% | 37.0 | 2025-11-24 |
| 24 | Gemini 2.0 Flash | 84.1% | 43.5 | 2025-11-24 |
| 25 | Gemini 2.5 Pro | 84.0% | 54.5 | 2025-11-24 |
| 26 | Claude Sonnet 4thinking | 84.0% | 53.3 | 2025-11-24 |
| 27 | Claude Sonnet 4 | 83.9% | 50.5 | 2025-11-24 |
| 28 | GPT-4.1 | 83.8% | 48.5 | 2025-11-24 |
| 29 | Gemini 1.5 Pro 002 | 83.7% | 41.5 | 2025-11-24 |
| 30 | gpt-oss-120bopen ↗ | 83.6% | 45.5 | 2025-11-24 |
| 31 |  Qwen3 235B A22B 2507open ↗ | 83.5% | 50.9 | 2025-11-24 |
| 32 | Claude 3.7 Sonnet | 83.4% | 49.9 | 2025-11-24 |
| 33 |  DeepSeek V3open ↗ | 83.2% | 44.5 | 2025-11-24 |
| 34 | Gemini 1.5 Flash 002 | 83.1% | 38.3 | 2025-11-24 |
| 35 | Gemini 2.0 Flash-Lite | 82.4% | 48.4 | 2025-11-24 |
| 36 | Palmyra X5 | 82.3% | – | 2025-11-24 |
| 37 | Llama 3.1 70Bopen ↗ | 82.1% | 40.9 | 2025-11-24 |
| 38 | Llama 4 Scoutopen ↗ | 81.8% | 38.6 | 2025-11-24 |
| 39 | GPT-4o | 81.7% | 43.3 | 2025-11-24 |
| 40 | Qwen3 235B A22Bopen ↗ | 81.6% | 50.0 | 2025-11-24 |
| 41 |  Nova Pro | 81.5% | 40.9 | 2025-11-24 |
| 42 |  GLM-4.5-Airopen ↗ | 81.2% | 46.8 | 2025-11-24 |
| 43 | Llama 3.1 405Bopen ↗ | 81.1% | 41.9 | 2025-11-24 |
| 44 | Qwen3-Next 80B A3Bthinkingopen ↗ | 81.0% | 52.3 | 2025-11-24 |
| 45 | Gemini 2.5 Flash-Lite | 81.0% | 48.3 | 2025-11-24 |
| 46 | Qwen2.5 72Bopen ↗ | 80.6% | 43.2 | 2025-11-24 |
| 47 | Nova Premier | 80.3% | 46.1 | 2025-11-24 |
| 48 | Claude Haiku 4.5 | 80.1% | 47.7 | 2025-11-24 |
| 49 | Palmyra Fin | 79.3% | – | 2025-11-24 |
| 50 | Claude 3.5 Haiku | 79.2% | 38.8 | 2025-11-24 |
| 51 | DeepSeek R1 0528open ↗ | 78.4% | 49.4 | 2025-11-24 |
| 52 | GPT-4o mini | 78.2% | 35.4 | 2025-11-24 |
| 53 |  Olmo 2 32B March 2025 | 78.0% | – | 2025-11-24 |
| 54 | Nova Lite | 77.6% | 38.3 | 2025-11-24 |
| 55 | Palmyra Med | 76.7% | – | 2025-11-24 |
| 56 | Nova Micro | 76.0% | 39.2 | 2025-11-24 |
| 57 | Mistral Small 3.1open | 75.0% | 38.5 | 2025-11-24 |
| 58 | Llama 3.1 8Bopen ↗ | 74.3% | 33.7 | 2025-11-24 |
| 59 | Qwen2.5 7B Instructopen ↗ | 74.1% | 39.9 | 2025-11-24 |
| 60 | gpt-oss-20bopen ↗ | 73.2% | 44.1 | 2025-11-24 |
More instruction following benchmarks · All LLM benchmarks
Related benchmarks
Cite as: BenchLeader, “IFEval (HELM) leaderboard”, data as of 29 Oct 2026, https://www.benchleader.com/benchmarks/helm_ifeval.
IFEval (HELM): questions
- What does IFEval (HELM) measure?
- IFEval gives the model writing tasks with verifiable instructions attached: respond in exactly three paragraphs, avoid the letter e, end with a given phrase, and so on. Each instruction can be checked by a program, so no judge model is needed. Scores are reported in percent of tasks solved; higher is better.
- What is the top score on IFEval (HELM)?
- As of 29 Oct 2026, the highest score on IFEval (HELM) is 95.1% by Grok 3 mini (result dated 24 Nov 2025), ahead of Grok 4 at 94.9%. Scores come from HELM Capabilities and are re-read daily.
- Which AI model leads IFEval (HELM)?
- Grok 3 mini leads IFEval (HELM) with 95.1% as of 29 Oct 2026, ahead of Grok 4 at 94.9% (result dated 24 Nov 2025).
- How many models have IFEval (HELM) results?
- 68 model configurations have a IFEval (HELM) result on BenchLeader, all taken from HELM Capabilities.
- Who runs IFEval (HELM) and how often is it updated?
- IFEval (HELM) is published by HELM Capabilities. BenchLeader re-reads the published results once a day and records the date each result was published.
- Does IFEval (HELM) count toward the BenchLeader Index?
- Yes. IFEval (HELM) contributes to the instruction following category of the BenchLeader Index, normalised so that 50 is the average of the evaluated models.