WildBench (HELM)
As of 29 Oct 2026, the top score on WildBench (HELM) is 86.6% by Qwen3 235B A22B 2507 (Alibaba), ahead of GPT-5.1 at 86.3%, across 68 model configurations with a published result (HELM Capabilities, result dated 24 Nov 2025). WildBench draws real tasks from users' conversations with chatbots, from coding help to creative writing, and grades responses with a judge model following task-specific checklists. WildBench (HELM) counts toward the BenchLeader instruction following score.
Key facts (as of 29 Oct 2026)
- Qwen3 235B A22B 2507 holds the top WildBench (HELM) score at 86.6% (HELM Capabilities, 24 Nov 2025).
- GPT-5.1 is second on WildBench (HELM) at 86.3% (HELM Capabilities, 24 Nov 2025).
- WildBench (HELM): Real user tasks graded by a rubric-guided judge. Run by Stanford CRFM's HELM (HELM Capabilities).
- 68 model configurations have a published WildBench (HELM) result on BenchLeader (BenchLeader, 29 Oct 2026).
- WildBench (HELM) feeds the instruction following category of the BenchLeader Index (BenchLeader methodology).
Real user tasks graded by a rubric-guided judge. Run by Stanford CRFM's HELM.
- Published by
- HELM Capabilities
- Category
- Instruction following
- Index weight
- 0.5
- 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
WildBench draws real tasks from users' conversations with chatbots, from coding help to creative writing, and grades responses with a judge model following task-specific checklists.
How it is scored
WB Score, a 0 to 100 rating from the checklist-guided judge, run by Stanford CRFM's HELM.
What to keep in mind
Judged by another model, so it inherits that judge's preferences; BenchLeader gives it half weight for that reason.
- 1Qwen3 235B A22B 250786.6%
- 2GPT-5.186.3%
- 3Kimi K286.2%
- 4o386.1%
- 5Gemini 3 Pro85.9%
- 6GPT-585.7%
- 7Gemini 2.5 Pro85.7%
- 8GPT-5 mini85.5%
- 9o4-mini85.4%
- 10Claude Sonnet 4.585.4%
- 11GPT-4.185.4%
- 12Claude Opus 4 (thinking)85.2%
- 13Grok 384.9%
- 14gpt-oss-120b84.5%
- 15Claude Haiku 4.583.9%
| # | ||||
|---|---|---|---|---|
| 1 |  Qwen3 235B A22B 2507open ↗ | 86.6% | 50.9 | 2025-11-24 |
| 2 |  GPT-5.1 | 86.3% | 56.8 | 2025-11-24 |
| 3 |  Kimi K2open ↗ | 86.2% | 51.0 | 2025-11-24 |
| 4 | o3 | 86.1% | 61.0 | 2025-11-24 |
| 5 |  Gemini 3 Pro | 85.9% | 60.2 | 2025-11-24 |
| 6 | GPT-5 | 85.7% | 57.7 | 2025-11-24 |
| 7 | Gemini 2.5 Pro | 85.7% | 54.5 | 2025-11-24 |
| 8 | GPT-5 mini | 85.5% | 56.4 | 2025-11-24 |
| 9 | o4-mini | 85.4% | 56.9 | 2025-11-24 |
| 10 |  Claude Sonnet 4.5 | 85.4% | 53.7 | 2025-11-24 |
| 11 | GPT-4.1 | 85.4% | 48.5 | 2025-11-24 |
| 12 | Claude Opus 4thinking | 85.2% | 55.2 | 2025-11-24 |
| 13 |  Grok 3 | 84.9% | 52.5 | 2025-11-24 |
| 14 | gpt-oss-120bopen ↗ | 84.5% | 45.5 | 2025-11-24 |
| 15 | Claude Haiku 4.5 | 83.9% | 47.7 | 2025-11-24 |
| 16 | Claude Sonnet 4thinking | 83.8% | 53.3 | 2025-11-24 |
| 17 | GPT-4.1 mini | 83.8% | 44.8 | 2025-11-24 |
| 18 | Claude Opus 4 | 83.3% | 52.6 | 2025-11-24 |
| 19 |  DeepSeek V3open ↗ | 83.1% | 44.5 | 2025-11-24 |
| 20 | Qwen3 235B A22Bopen ↗ | 82.8% | 50.0 | 2025-11-24 |
| 21 | DeepSeek R1 0528open ↗ | 82.8% | 49.4 | 2025-11-24 |
| 22 | GPT-4o | 82.8% | 43.3 | 2025-11-24 |
| 23 | Claude Sonnet 4 | 82.5% | 50.5 | 2025-11-24 |
| 24 | Gemini 2.5 Flash-Lite | 81.8% | 48.3 | 2025-11-24 |
| 25 | Gemini 2.5 Flash | 81.7% | 52.7 | 2025-11-24 |
| 26 | Claude 3.7 Sonnet | 81.4% | 49.9 | 2025-11-24 |
| 27 | Gemini 1.5 Pro 002 | 81.3% | 41.5 | 2025-11-24 |
| 28 | GPT-4.1 nano | 81.1% | 37.0 | 2025-11-24 |
| 29 | Qwen3-Next 80B A3Bthinkingopen ↗ | 80.7% | 52.3 | 2025-11-24 |
| 30 | GPT-5 nano | 80.6% | 55.9 | 2025-11-24 |
| 31 | Qwen2.5 72Bopen ↗ | 80.2% | 43.2 | 2025-11-24 |
| 32 |  Palmyra X 004 | 80.2% | – | 2025-11-24 |
| 33 |  Mistral Large 2.1open | 80.1% | 42.6 | 2025-11-24 |
| 34 | Gemini 2.0 Flash | 80.0% | 43.5 | 2025-11-24 |
| 35 |  Llama 4 Maverickopen ↗ | 80.0% | 42.2 | 2025-11-24 |
| 36 | Grok 4 | 79.7% | 58.4 | 2025-11-24 |
| 37 | Claude 3.5 Sonnet | 79.2% | 47.1 | 2025-11-24 |
| 38 | Gemini 1.5 Flash 002 | 79.2% | 38.3 | 2025-11-24 |
| 39 | GPT-4o mini | 79.1% | 35.4 | 2025-11-24 |
| 40 | Gemini 2.0 Flash-Lite | 79.0% | 48.4 | 2025-11-24 |
| 41 |  GLM-4.5-Airopen ↗ | 78.9% | 46.8 | 2025-11-24 |
| 42 |  Nova Premier | 78.8% | 46.1 | 2025-11-24 |
| 43 | Mistral Small 3.1open | 78.8% | 38.5 | 2025-11-24 |
| 44 | Llama 3.1 405Bopen ↗ | 78.3% | 41.9 | 2025-11-24 |
| 45 | Palmyra Fin | 78.3% | – | 2025-11-24 |
| 46 | Palmyra X5 | 78.0% | – | 2025-11-24 |
| 47 | Llama 4 Scoutopen ↗ | 77.9% | 38.6 | 2025-11-24 |
| 48 | Nova Pro | 77.7% | 40.9 | 2025-11-24 |
| 49 | Claude 3.5 Haiku | 76.0% | 38.8 | 2025-11-24 |
| 50 | Llama 3.1 70Bopen ↗ | 75.8% | 40.9 | 2025-11-24 |
| 51 | Nova Lite | 75.0% | 38.3 | 2025-11-24 |
| 52 | Nova Micro | 74.3% | 39.2 | 2025-11-24 |
| 53 |  Granite 3.3 8Bopen ↗ | 74.1% | – | 2025-11-24 |
| 54 | Granite 4.0 Small | 73.9% | – | 2025-11-24 |
| 55 | gpt-oss-20bopen ↗ | 73.7% | 44.1 | 2025-11-24 |
| 56 |  Olmo 2 32B March 2025 | 73.4% | – | 2025-11-24 |
| 57 | Qwen2.5 7B Instructopen ↗ | 73.1% | 39.9 | 2025-11-24 |
| 58 | Mixtral 8x22B Instructopen ↗ | 71.1% | 37.5 | 2025-11-24 |
| 59 | Olmo 2 13B November 2024 | 68.9% | – | 2025-11-24 |
| 60 | Llama 3.1 8Bopen ↗ | 68.6% | 33.7 | 2025-11-24 |
More instruction following benchmarks · All LLM benchmarks
Related benchmarks
Top head-to-heads on WildBench (HELM)
Cite as: BenchLeader, “WildBench (HELM) leaderboard”, data as of 29 Oct 2026, https://www.benchleader.com/benchmarks/helm_wildbench.
WildBench (HELM): questions
- What does WildBench (HELM) measure?
- WildBench draws real tasks from users' conversations with chatbots, from coding help to creative writing, and grades responses with a judge model following task-specific checklists. Scores are reported in percent of tasks solved; higher is better.
- What is the top score on WildBench (HELM)?
- As of 29 Oct 2026, the highest score on WildBench (HELM) is 86.6% by Qwen3 235B A22B 2507 (result dated 24 Nov 2025), ahead of GPT-5.1 at 86.3%. Scores come from HELM Capabilities and are re-read daily.
- Which AI model leads WildBench (HELM)?
- Qwen3 235B A22B 2507 leads WildBench (HELM) with 86.6% as of 29 Oct 2026, ahead of GPT-5.1 at 86.3% (result dated 24 Nov 2025).
- How many models have WildBench (HELM) results?
- 68 model configurations have a WildBench (HELM) result on BenchLeader, all taken from HELM Capabilities.
- Who runs WildBench (HELM) and how often is it updated?
- WildBench (HELM) is published by HELM Capabilities. BenchLeader re-reads the published results once a day and records the date each result was published.
- Does WildBench (HELM) count toward the BenchLeader Index?
- Yes. WildBench (HELM) contributes to the instruction following category of the BenchLeader Index, normalised so that 50 is the average of the evaluated models.