Knowledge benchmark

# GPQA (HELM) leaderboard

> GPQA (HELM) results for 57 AI models, led by Gemini 3 Pro at 80.3%. What the benchmark measures, who runs it, and a source for every score.
- Canonical page: https://noometry.com/benchmarks/helm-gpqa
- Last updated: 2026-10-10
- Title: GPQA (HELM) Leaderboard (October 2026): Scores by Model

As of October 2026, Gemini 3 Pro has the highest published GPQA (HELM) score on Noometry at 80.3%, out of 57 models with results.

Last verified October 10, 2026

## About GPQA (HELM)

Graduate-level biology, physics and chemistry questions written to be hard to look up, run by HELM Capabilities with chain-of-thought.

- **Category:** [Knowledge](https://noometry.com/best/knowledge)
- **Introduced:** 2023
- **Format:** 4-option multiple choice
- **Unit:** Percent (random guessing ≈ 25%)
- **Official site:** [crfm.stanford.edu](https://crfm.stanford.edu/helm/capabilities/latest/)

## Top 15 models

Top models on GPQA (HELM)

1.  Gemini 3 Pro 80.3%
2.  GPT-5 79.2%
3.  GPT-5 Mini 75.6%
4.  o3 75.3%
5.  Gemini 2.5 Pro 74.9%
6.  o4-mini 73.5%
7.  Grok 4 72.7%
8.  Qwen3 235B-A22B 72.7%
9.  Claude Opus 4 70.8%
10.  Claude Sonnet 4 70.6%
11.  Claude Sonnet 4.5 68.6%
12.  gpt-oss-120b 68.4%
13.  GPT-5 Nano 67.9%
14.  Grok-3 mini 67.5%
15.  DeepSeek-R1 66.6%
16.  606570758085

Sponsored placements are available on pages like this one. [Advertise on Noometry](https://noometry.com/advertise)

## All results

GPQA (HELM) results by model
| # | Model | Provider | Score | Setting | Source | Date |
| --- | --- | --- | --- | --- | --- | --- |
| 1 | [Gemini 3 Pro](https://noometry.com/models/gemini-3-pro) |  [![](/logos/google.svg) Google](https://noometry.com/providers/google) | 80.3% |  | [HELM Capabilities](https://crfm.stanford.edu/helm/capabilities/latest/) |  |
| 2 | [GPT-5](https://noometry.com/models/gpt-5) | [OpenAI](https://noometry.com/providers/openai) | 79.2% |  | [HELM Capabilities](https://crfm.stanford.edu/helm/capabilities/latest/) |  |
| 3 | [GPT-5 Mini](https://noometry.com/models/gpt-5-mini) | [OpenAI](https://noometry.com/providers/openai) | 75.6% |  | [HELM Capabilities](https://crfm.stanford.edu/helm/capabilities/latest/) |  |
| 4 | [o3](https://noometry.com/models/o3) | [OpenAI](https://noometry.com/providers/openai) | 75.3% |  | [HELM Capabilities](https://crfm.stanford.edu/helm/capabilities/latest/) |  |
| 5 | [Gemini 2.5 Pro](https://noometry.com/models/gemini-2-5-pro) |  [![](/logos/google.svg) Google](https://noometry.com/providers/google) | 74.9% |  | [HELM Capabilities](https://crfm.stanford.edu/helm/capabilities/latest/) |  |
| 6 | [o4-mini](https://noometry.com/models/o4-mini) | [OpenAI](https://noometry.com/providers/openai) | 73.5% |  | [HELM Capabilities](https://crfm.stanford.edu/helm/capabilities/latest/) |  |
| 7 | [Grok 4](https://noometry.com/models/grok-4) | [xAI](https://noometry.com/providers/xai) | 72.7% |  | [HELM Capabilities](https://crfm.stanford.edu/helm/capabilities/latest/) |  |
| 8 | [Qwen3 235B-A22B](https://noometry.com/models/qwen3-235b-a22b) |  [![](/logos/alibaba.svg) Alibaba (Qwen)](https://noometry.com/providers/alibaba) | 72.7% |  | [HELM Capabilities](https://crfm.stanford.edu/helm/capabilities/latest/) |  |
| 9 | [Claude Opus 4](https://noometry.com/models/claude-opus-4) | [Anthropic](https://noometry.com/providers/anthropic) | 70.8% |  | [HELM Capabilities](https://crfm.stanford.edu/helm/capabilities/latest/) |  |
| 10 | [Claude Sonnet 4](https://noometry.com/models/claude-sonnet-4) | [Anthropic](https://noometry.com/providers/anthropic) | 70.6% |  | [HELM Capabilities](https://crfm.stanford.edu/helm/capabilities/latest/) |  |
| 11 | [Claude Sonnet 4.5](https://noometry.com/models/claude-sonnet-4-5) | [Anthropic](https://noometry.com/providers/anthropic) | 68.6% |  | [HELM Capabilities](https://crfm.stanford.edu/helm/capabilities/latest/) |  |
| 12 | [gpt-oss-120b](https://noometry.com/models/gpt-oss-120b) | [OpenAI](https://noometry.com/providers/openai) | 68.4% |  | [HELM Capabilities](https://crfm.stanford.edu/helm/capabilities/latest/) |  |
| 13 | [GPT-5 Nano](https://noometry.com/models/gpt-5-nano) | [OpenAI](https://noometry.com/providers/openai) | 67.9% |  | [HELM Capabilities](https://crfm.stanford.edu/helm/capabilities/latest/) |  |
| 14 | [Grok-3 mini](https://noometry.com/models/grok-3-mini) | [xAI](https://noometry.com/providers/xai) | 67.5% |  | [HELM Capabilities](https://crfm.stanford.edu/helm/capabilities/latest/) |  |
| 15 | [DeepSeek-R1](https://noometry.com/models/deepseek-r1) |  [![](/logos/deepseek.svg) DeepSeek](https://noometry.com/providers/deepseek) | 66.6% |  | [HELM Capabilities](https://crfm.stanford.edu/helm/capabilities/latest/) |  |
| 16 | [GPT-4.1](https://noometry.com/models/gpt-4-1) | [OpenAI](https://noometry.com/providers/openai) | 65.9% |  | [HELM Capabilities](https://crfm.stanford.edu/helm/capabilities/latest/) |  |
| 17 | [Kimi K2 (Jul 2025)](https://noometry.com/models/kimi-k2) | [Moonshot AI](https://noometry.com/providers/moonshot) | 65.3% |  | [HELM Capabilities](https://crfm.stanford.edu/helm/capabilities/latest/) |  |
| 18 | [Grok 3](https://noometry.com/models/grok-3) | [xAI](https://noometry.com/providers/xai) | 65% |  | [HELM Capabilities](https://crfm.stanford.edu/helm/capabilities/latest/) |  |
| 19 | [Llama 4 Maverick](https://noometry.com/models/llama-4-maverick) |  [![](/logos/meta.svg) Meta](https://noometry.com/providers/meta) | 65% |  | [HELM Capabilities](https://crfm.stanford.edu/helm/capabilities/latest/) |  |
| 20 | [Qwen3-Next 80B-A3B Instruct](https://noometry.com/models/qwen3-next-80b-a3b-instruct) |  [![](/logos/alibaba.svg) Alibaba (Qwen)](https://noometry.com/providers/alibaba) | 63% |  | [HELM Capabilities](https://crfm.stanford.edu/helm/capabilities/latest/) |  |
| 21 | [GPT-4.1 mini](https://noometry.com/models/gpt-4-1-mini) | [OpenAI](https://noometry.com/providers/openai) | 61.4% |  | [HELM Capabilities](https://crfm.stanford.edu/helm/capabilities/latest/) |  |
| 22 | [Claude 3.7 Sonnet](https://noometry.com/models/claude-3-7-sonnet) | [Anthropic](https://noometry.com/providers/anthropic) | 60.8% |  | [HELM Capabilities](https://crfm.stanford.edu/helm/capabilities/latest/) |  |
| 23 | [Claude Haiku 4.5](https://noometry.com/models/claude-haiku-4-5) | [Anthropic](https://noometry.com/providers/anthropic) | 60.5% |  | [HELM Capabilities](https://crfm.stanford.edu/helm/capabilities/latest/) |  |
| 24 | [GLM-4.5-Air](https://noometry.com/models/glm-4-5-air) | [Z.ai (Zhipu)](https://noometry.com/providers/zai) | 59.4% |  | [HELM Capabilities](https://crfm.stanford.edu/helm/capabilities/latest/) |  |
| 25 | [gpt-oss-20b](https://noometry.com/models/gpt-oss-20b) | [OpenAI](https://noometry.com/providers/openai) | 59.4% |  | [HELM Capabilities](https://crfm.stanford.edu/helm/capabilities/latest/) |  |
| 26 | [Claude 3.5 Sonnet](https://noometry.com/models/claude-3-5-sonnet) | [Anthropic](https://noometry.com/providers/anthropic) | 56.5% |  | [HELM Capabilities](https://crfm.stanford.edu/helm/capabilities/latest/) |  |
| 27 | [Gemini 2.0 Flash (Feb 2025)](https://noometry.com/models/gemini-2-0-flash) |  [![](/logos/google.svg) Google](https://noometry.com/providers/google) | 55.6% |  | [HELM Capabilities](https://crfm.stanford.edu/helm/capabilities/latest/) |  |
| 28 | [DeepSeek-V3](https://noometry.com/models/deepseek-v3) |  [![](/logos/deepseek.svg) DeepSeek](https://noometry.com/providers/deepseek) | 53.8% |  | [HELM Capabilities](https://crfm.stanford.edu/helm/capabilities/latest/) |  |
| 29 | [Gemini 1.5 Pro (May 2024)](https://noometry.com/models/gemini-1-5-pro) |  [![](/logos/google.svg) Google](https://noometry.com/providers/google) | 53.4% |  | [HELM Capabilities](https://crfm.stanford.edu/helm/capabilities/latest/) |  |
| 30 | [Llama 3.1-405B](https://noometry.com/models/llama-3-1-405b) |  [![](/logos/meta.svg) Meta](https://noometry.com/providers/meta) | 52.2% |  | [HELM Capabilities](https://crfm.stanford.edu/helm/capabilities/latest/) |  |
| 31 | [GPT-4o](https://noometry.com/models/gpt-4o) | [OpenAI](https://noometry.com/providers/openai) | 52% |  | [HELM Capabilities](https://crfm.stanford.edu/helm/capabilities/latest/) |  |
| 32 | [Nova Premier 1.0](https://noometry.com/models/nova-premier-1-0) | [Amazon](https://noometry.com/providers/amazon) | 51.8% |  | [HELM Capabilities](https://crfm.stanford.edu/helm/capabilities/latest/) |  |
| 33 | [GPT-4.1 nano](https://noometry.com/models/gpt-4-1-nano) | [OpenAI](https://noometry.com/providers/openai) | 50.7% |  | [HELM Capabilities](https://crfm.stanford.edu/helm/capabilities/latest/) |  |
| 34 | [Llama 4 Scout](https://noometry.com/models/llama-4-scout) |  [![](/logos/meta.svg) Meta](https://noometry.com/providers/meta) | 50.7% |  | [HELM Capabilities](https://crfm.stanford.edu/helm/capabilities/latest/) |  |
| 35 | [Gemini 2.0 Flash-Lite](https://noometry.com/models/gemini-2-0-flash-lite) |  [![](/logos/google.svg) Google](https://noometry.com/providers/google) | 50% |  | [HELM Capabilities](https://crfm.stanford.edu/helm/capabilities/latest/) |  |
| 36 | [Amazon Nova Pro](https://noometry.com/models/amazon-nova-pro) | [Amazon](https://noometry.com/providers/amazon) | 44.6% |  | [HELM Capabilities](https://crfm.stanford.edu/helm/capabilities/latest/) |  |
| 37 | [GPT-5.1](https://noometry.com/models/gpt-5-1) | [OpenAI](https://noometry.com/providers/openai) | 44.2% |  | [HELM Capabilities](https://crfm.stanford.edu/helm/capabilities/latest/) |  |
| 38 | [Gemini 1.5 Flash (May 2024)](https://noometry.com/models/gemini-1-5-flash) |  [![](/logos/google.svg) Google](https://noometry.com/providers/google) | 43.7% |  | [HELM Capabilities](https://crfm.stanford.edu/helm/capabilities/latest/) |  |
| 39 | [Mistral Large](https://noometry.com/models/mistral-large) |  [![](/logos/mistral.svg) Mistral AI](https://noometry.com/providers/mistral) | 43.5% |  | [HELM Capabilities](https://crfm.stanford.edu/helm/capabilities/latest/) |  |
| 40 | [Llama 3.1-70B](https://noometry.com/models/llama-3-1-70b) |  [![](/logos/meta.svg) Meta](https://noometry.com/providers/meta) | 42.6% |  | [HELM Capabilities](https://crfm.stanford.edu/helm/capabilities/latest/) |  |
| 41 | [Qwen2.5 72B Instruct](https://noometry.com/models/qwen2-5-72b-instruct) |  [![](/logos/alibaba.svg) Alibaba (Qwen)](https://noometry.com/providers/alibaba) | 42.6% |  | [HELM Capabilities](https://crfm.stanford.edu/helm/capabilities/latest/) |  |
| 42 | [Amazon Nova Lite](https://noometry.com/models/amazon-nova-lite) | [Amazon](https://noometry.com/providers/amazon) | 39.7% |  | [HELM Capabilities](https://crfm.stanford.edu/helm/capabilities/latest/) |  |
| 43 | [Mistral Small 3.1](https://noometry.com/models/mistral-small-3-1) |  [![](/logos/mistral.svg) Mistral AI](https://noometry.com/providers/mistral) | 39.2% |  | [HELM Capabilities](https://crfm.stanford.edu/helm/capabilities/latest/) |  |
| 44 | [Gemini 2.5 Flash](https://noometry.com/models/gemini-2-5-flash) |  [![](/logos/google.svg) Google](https://noometry.com/providers/google) | 39% |  | [HELM Capabilities](https://crfm.stanford.edu/helm/capabilities/latest/) |  |
| 45 | [Amazon Nova Micro](https://noometry.com/models/amazon-nova-micro) | [Amazon](https://noometry.com/providers/amazon) | 38.3% |  | [HELM Capabilities](https://crfm.stanford.edu/helm/capabilities/latest/) |  |
| 46 | [Granite 4.0 H Small](https://noometry.com/models/ibm-granite-h-small) | [IBM](https://noometry.com/providers/ibm) | 38.3% |  | [HELM Capabilities](https://crfm.stanford.edu/helm/capabilities/latest/) |  |
| 47 | [GPT-4o mini](https://noometry.com/models/gpt-4o-mini) | [OpenAI](https://noometry.com/providers/openai) | 36.8% |  | [HELM Capabilities](https://crfm.stanford.edu/helm/capabilities/latest/) |  |
| 48 | [Claude 3.5 Haiku](https://noometry.com/models/claude-3-5-haiku) | [Anthropic](https://noometry.com/providers/anthropic) | 36.3% |  | [HELM Capabilities](https://crfm.stanford.edu/helm/capabilities/latest/) |  |
| 49 | [Qwen2.5 7B Instruct](https://noometry.com/models/qwen2-5-7b-instruct) |  [![](/logos/alibaba.svg) Alibaba (Qwen)](https://noometry.com/providers/alibaba) | 34.1% |  | [HELM Capabilities](https://crfm.stanford.edu/helm/capabilities/latest/) |  |
| 50 | [Mixtral 8x22B](https://noometry.com/models/mixtral-8x22b) |  [![](/logos/mistral.svg) Mistral AI](https://noometry.com/providers/mistral) | 33.4% |  | [HELM Capabilities](https://crfm.stanford.edu/helm/capabilities/latest/) |  |
| 51 | [OLMo 2 Furious 13B](https://noometry.com/models/olmo-2-furious-13b) |  [![](/logos/ai2.svg) Allen Institute for AI (Ai2)](https://noometry.com/providers/ai2) | 31.6% |  | [HELM Capabilities](https://crfm.stanford.edu/helm/capabilities/latest/) |  |
| 52 | [Gemini 2.5 Flash-Lite](https://noometry.com/models/gemini-2-5-flash-lite) |  [![](/logos/google.svg) Google](https://noometry.com/providers/google) | 30.9% |  | [HELM Capabilities](https://crfm.stanford.edu/helm/capabilities/latest/) |  |
| 53 | [Granite 4.0 Micro](https://noometry.com/models/granite-4-0-micro) | [IBM](https://noometry.com/providers/ibm) | 30.7% |  | [HELM Capabilities](https://crfm.stanford.edu/helm/capabilities/latest/) |  |
| 54 | [Mistral](https://noometry.com/models/mistral) |  [![](/logos/mistral.svg) Mistral AI](https://noometry.com/providers/mistral) | 30.3% |  | [HELM Capabilities](https://crfm.stanford.edu/helm/capabilities/latest/) |  |
| 55 | [Mixtral 8x7B](https://noometry.com/models/mixtral-8x7b) |  [![](/logos/mistral.svg) Mistral AI](https://noometry.com/providers/mistral) | 29.6% |  | [HELM Capabilities](https://crfm.stanford.edu/helm/capabilities/latest/) |  |
| 56 | [Olmo 2 0325 32b Instruct](https://noometry.com/models/olmo-2-0325-32b-instruct) |  [![](/logos/ai2.svg) Allen Institute for AI (Ai2)](https://noometry.com/providers/ai2) | 28.7% |  | [HELM Capabilities](https://crfm.stanford.edu/helm/capabilities/latest/) |  |
| 57 | [Llama 3.1-8B](https://noometry.com/models/llama-3-1-8b) |  [![](/logos/meta.svg) Meta](https://noometry.com/providers/meta) | 24.7% |  | [HELM Capabilities](https://crfm.stanford.edu/helm/capabilities/latest/) |  |

## Compare the leaders

-   [Gemini 3 Pro vs GPT-5](https://noometry.com/compare/gemini-3-pro-vs-gpt-5)
-   [Gemini 3 Pro vs GPT-5 Mini](https://noometry.com/compare/gemini-3-pro-vs-gpt-5-mini)
-   [Gemini 3 Pro vs o3](https://noometry.com/compare/gemini-3-pro-vs-o3)
-   [Gemini 3 Pro vs Gemini 2.5 Pro](https://noometry.com/compare/gemini-2-5-pro-vs-gemini-3-pro)
-   [GPT-5 vs GPT-5 Mini](https://noometry.com/compare/gpt-5-vs-gpt-5-mini)
-   [GPT-5 vs o3](https://noometry.com/compare/gpt-5-vs-o3)

## Other knowledge benchmarks

-   [GPQA Diamond](https://noometry.com/benchmarks/gpqa-diamond)
-   [Humanity's Last Exam](https://noometry.com/benchmarks/hle)
-   [SimpleQA Verified](https://noometry.com/benchmarks/simpleqa-verified)
-   [MMLU-Pro](https://noometry.com/benchmarks/mmlu-pro)
-   [Confabulations](https://noometry.com/benchmarks/confabulations)
-   [Vectara Hallucination Rate](https://noometry.com/benchmarks/vectara-hallucination)
-   [LMArena Expert](https://noometry.com/benchmarks/arena-expert)
-   [ARC (AI2) Challenge](https://noometry.com/benchmarks/arc-challenge) (reference)
-   [BoolQ](https://noometry.com/benchmarks/boolq) (reference)
-   [MMLU](https://noometry.com/benchmarks/mmlu) (reference)
-   [OpenBookQA](https://noometry.com/benchmarks/openbookqa) (reference)
-   [TriviaQA](https://noometry.com/benchmarks/triviaqa) (reference)

## Frequently asked questions

### What does GPQA (HELM) measure?

Graduate-level biology, physics and chemistry questions written to be hard to look up, run by HELM Capabilities with chain-of-thought.

### Which model has the highest GPQA (HELM) score?

As of October 2026, Gemini 3 Pro has the highest published GPQA (HELM) score on Noometry at 80.3%, out of 57 models with results.

### What is the best open-weight model on GPQA (HELM)?

Qwen3 235B-A22B has the highest GPQA (HELM) accuracy among open-weight models at 72.7%, ranking 8 of 57 overall.

### Cite this page

Noometry. (2026). GPQA (HELM) leaderboard. Retrieved October 10, 2026, from https://noometry.com/benchmarks/helm-gpqa

Quote Noometry with a link back to this page. It is also available in [Markdown](https://noometry.com/md/benchmarks/helm-gpqa.md).