Agentic & Tool Use benchmark # BALROG leaderboard > BALROG results for 35 AI models, led by GPT-6 Astra at 68.3%. What the benchmark measures, who runs it, and a source for every score. - Canonical page: https://noometry.com/benchmarks/balrog - Last updated: 2026-10-10 - Title: BALROG Leaderboard (October 2026): Scores by Model As of October 2026, GPT-6 Astra has the highest published BALROG score on Noometry at 68.3%, out of 35 models with results. Last verified October 10, 2026 ## About BALROG Long-horizon game environments, from BabyAI to NetHack, that test planning and exploration. - **Category:** [Agentic & Tool Use](https://noometry.com/best/agentic) - **Introduced:** 2024 - **Format:** Games - **Unit:** Percent (random guessing ≈ 0%) - **Official site:** [balrogai.com](https://balrogai.com) ## Top 15 models Top models on BALROG 1. GPT-6 Astra 68.3% 2. Claude Opus 5 63.4% 3. GPT-5.6 Sol 60% 4. Gemini 3 Pro 58.1% 5. Gemini 3.1 Pro Preview 57% 6. GPT-5.6 Terra 53.2% 7. Gemini 3 Flash Preview 48.1% 8. GPT-5.6 Luna 45.6% 9. Grok 4 43.6% 10. Claude Opus 4.5 43.5% 11. Gemini 2.5 Pro 43.3% 12. DeepSeek-R1 34.9% 13. Gemini 2.5 Flash 33.5% 14. GPT-5 32.8% 15. Claude 3.5 Sonnet 32.6% 16. 020406080 Sponsored placements are available on pages like this one. [Advertise on Noometry](https://noometry.com/advertise) ## All results BALROG results by model | # | Model | Provider | Score | Setting | Source | Date | | --- | --- | --- | --- | --- | --- | --- | | 1 | [GPT-6 Astra](https://noometry.com/models/gpt-6-astra) | [OpenAI](https://noometry.com/providers/openai) | 68.3% | max | [Epoch AI](https://epoch.ai/benchmarks) | | | 2 | [Claude Opus 5](https://noometry.com/models/claude-opus-5) | [Anthropic](https://noometry.com/providers/anthropic) | 63.4% | max | [Epoch AI](https://epoch.ai/benchmarks) | | | 3 | [GPT-5.6 Sol](https://noometry.com/models/gpt-5-6-sol) | [OpenAI](https://noometry.com/providers/openai) | 60% | max | [Epoch AI](https://epoch.ai/benchmarks) | | | 4 | [Gemini 3 Pro](https://noometry.com/models/gemini-3-pro) | [![](/logos/google.svg) Google](https://noometry.com/providers/google) | 58.1% | | [Epoch AI](https://epoch.ai/benchmarks) | | | 5 | [Gemini 3.1 Pro Preview](https://noometry.com/models/gemini-3-1-pro-preview) | [![](/logos/google.svg) Google](https://noometry.com/providers/google) | 57% | | [Epoch AI](https://epoch.ai/benchmarks) | | | 6 | [GPT-5.6 Terra](https://noometry.com/models/gpt-5-6-terra) | [OpenAI](https://noometry.com/providers/openai) | 53.2% | max | [Epoch AI](https://epoch.ai/benchmarks) | | | 7 | [Gemini 3 Flash Preview](https://noometry.com/models/gemini-3-flash-preview) | [![](/logos/google.svg) Google](https://noometry.com/providers/google) | 48.1% | | [Epoch AI](https://epoch.ai/benchmarks) | | | 8 | [GPT-5.6 Luna](https://noometry.com/models/gpt-5-6-luna) | [OpenAI](https://noometry.com/providers/openai) | 45.6% | max | [Epoch AI](https://epoch.ai/benchmarks) | | | 9 | [Grok 4](https://noometry.com/models/grok-4) | [xAI](https://noometry.com/providers/xai) | 43.6% | | [Epoch AI](https://epoch.ai/benchmarks) | | | 10 | [Claude Opus 4.5](https://noometry.com/models/claude-opus-4-5) | [Anthropic](https://noometry.com/providers/anthropic) | 43.5% | | [Epoch AI](https://epoch.ai/benchmarks) | | | 11 | [Gemini 2.5 Pro](https://noometry.com/models/gemini-2-5-pro) | [![](/logos/google.svg) Google](https://noometry.com/providers/google) | 43.3% | | [Epoch AI](https://epoch.ai/benchmarks) | | | 12 | [DeepSeek-R1](https://noometry.com/models/deepseek-r1) | [![](/logos/deepseek.svg) DeepSeek](https://noometry.com/providers/deepseek) | 34.9% | | [Epoch AI](https://epoch.ai/benchmarks) | | | 13 | [Gemini 2.5 Flash](https://noometry.com/models/gemini-2-5-flash) | [![](/logos/google.svg) Google](https://noometry.com/providers/google) | 33.5% | | [Epoch AI](https://epoch.ai/benchmarks) | | | 14 | [GPT-5](https://noometry.com/models/gpt-5) | [OpenAI](https://noometry.com/providers/openai) | 32.8% | minimal | [Epoch AI](https://epoch.ai/benchmarks) | | | 15 | [Claude 3.5 Sonnet](https://noometry.com/models/claude-3-5-sonnet) | [Anthropic](https://noometry.com/providers/anthropic) | 32.6% | | [Epoch AI](https://epoch.ai/benchmarks) | | | 16 | [GPT-4o](https://noometry.com/models/gpt-4o) | [OpenAI](https://noometry.com/providers/openai) | 32.3% | | [Epoch AI](https://epoch.ai/benchmarks) | | | 17 | [Claude Haiku 4.5](https://noometry.com/models/claude-haiku-4-5) | [Anthropic](https://noometry.com/providers/anthropic) | 31.2% | | [Epoch AI](https://epoch.ai/benchmarks) | | | 18 | [Grok 3](https://noometry.com/models/grok-3) | [xAI](https://noometry.com/providers/xai) | 29.5% | | [Epoch AI](https://epoch.ai/benchmarks) | | | 19 | [Reka Flash 3](https://noometry.com/models/reka-flash-3) | [Reka](https://noometry.com/providers/reka) | 29.2% | | [Epoch AI](https://epoch.ai/benchmarks) | | | 20 | [Llama 3.1-70B](https://noometry.com/models/llama-3-1-70b) | [![](/logos/meta.svg) Meta](https://noometry.com/providers/meta) | 27.9% | | [Epoch AI](https://epoch.ai/benchmarks) | | | 21 | [Llama 3.2 90B](https://noometry.com/models/llama-3-2-90b) | [![](/logos/meta.svg) Meta](https://noometry.com/providers/meta) | 27.3% | | [Epoch AI](https://epoch.ai/benchmarks) | | | 22 | [Llama-3.3-70B-Instruct](https://noometry.com/models/llama-3-3-70b-instruct) | [![](/logos/meta.svg) Meta](https://noometry.com/providers/meta) | 23% | | [Epoch AI](https://epoch.ai/benchmarks) | | | 23 | [Gemini 1.5 Pro (May 2024)](https://noometry.com/models/gemini-1-5-pro) | [![](/logos/google.svg) Google](https://noometry.com/providers/google) | 21% | | [Epoch AI](https://epoch.ai/benchmarks) | | | 24 | [DeepSeek-R1-Distill-Qwen-32B](https://noometry.com/models/deepseek-r1-distill-qwen-32b) | [![](/logos/deepseek.svg) DeepSeek](https://noometry.com/providers/deepseek) | 19.5% | | [Epoch AI](https://epoch.ai/benchmarks) | | | 25 | [Claude 3.5 Haiku](https://noometry.com/models/claude-3-5-haiku) | [Anthropic](https://noometry.com/providers/anthropic) | 19.3% | | [Epoch AI](https://epoch.ai/benchmarks) | | | 26 | [Mistral Nemo](https://noometry.com/models/mistral-nemo) | [![](/logos/mistral.svg) Mistral AI](https://noometry.com/providers/mistral) | 17.6% | | [Epoch AI](https://epoch.ai/benchmarks) | | | 27 | [GPT-4o mini](https://noometry.com/models/gpt-4o-mini) | [OpenAI](https://noometry.com/providers/openai) | 17.4% | | [Epoch AI](https://epoch.ai/benchmarks) | | | 28 | [Llama 3.2 11B](https://noometry.com/models/llama-3-2-11b) | [![](/logos/meta.svg) Meta](https://noometry.com/providers/meta) | 16.8% | | [Epoch AI](https://epoch.ai/benchmarks) | | | 29 | [Qwen2.5 72B Instruct](https://noometry.com/models/qwen2-5-72b-instruct) | [![](/logos/alibaba.svg) Alibaba (Qwen)](https://noometry.com/providers/alibaba) | 16.2% | | [Epoch AI](https://epoch.ai/benchmarks) | | | 30 | [Llama 3.1-8B](https://noometry.com/models/llama-3-1-8b) | [![](/logos/meta.svg) Meta](https://noometry.com/providers/meta) | 15.1% | | [Epoch AI](https://epoch.ai/benchmarks) | | | 31 | [Gemini 1.5 Flash (May 2024)](https://noometry.com/models/gemini-1-5-flash) | [![](/logos/google.svg) Google](https://noometry.com/providers/google) | 14.6% | | [Epoch AI](https://epoch.ai/benchmarks) | | | 32 | [Phi-4](https://noometry.com/models/phi-4) | [![](/logos/microsoft.svg) Microsoft](https://noometry.com/providers/microsoft) | 11.6% | | [Epoch AI](https://epoch.ai/benchmarks) | | | 33 | [Llama 3.2 3B](https://noometry.com/models/llama-3-2-3b) | [![](/logos/meta.svg) Meta](https://noometry.com/providers/meta) | 10.1% | | [Epoch AI](https://epoch.ai/benchmarks) | | | 34 | [Qwen2.5 7B Instruct](https://noometry.com/models/qwen2-5-7b-instruct) | [![](/logos/alibaba.svg) Alibaba (Qwen)](https://noometry.com/providers/alibaba) | 7.8% | | [Epoch AI](https://epoch.ai/benchmarks) | | | 35 | [Llama 3.2 1B](https://noometry.com/models/llama-3-2-1b) | [![](/logos/meta.svg) Meta](https://noometry.com/providers/meta) | 6.6% | | [Epoch AI](https://epoch.ai/benchmarks) | | ## Compare the leaders - [GPT-6 Astra vs Claude Opus 5](https://noometry.com/compare/claude-opus-5-vs-gpt-6-astra) - [GPT-6 Astra vs GPT-5.6 Sol](https://noometry.com/compare/gpt-5-6-sol-vs-gpt-6-astra) - [GPT-6 Astra vs Gemini 3 Pro](https://noometry.com/compare/gemini-3-pro-vs-gpt-6-astra) - [GPT-6 Astra vs Gemini 3.1 Pro Preview](https://noometry.com/compare/gemini-3-1-pro-preview-vs-gpt-6-astra) - [Claude Opus 5 vs GPT-5.6 Sol](https://noometry.com/compare/claude-opus-5-vs-gpt-5-6-sol) - [Claude Opus 5 vs Gemini 3 Pro](https://noometry.com/compare/claude-opus-5-vs-gemini-3-pro) ## Other agentic & tool use benchmarks - [Terminal-Bench](https://noometry.com/benchmarks/terminal-bench) - [APEX-Agents](https://noometry.com/benchmarks/apex-agents) - [Berkeley Function Calling Leaderboard](https://noometry.com/benchmarks/bfcl) - [OSWorld 2.0](https://noometry.com/benchmarks/osworld-2) - [GDPval](https://noometry.com/benchmarks/gdpval) - [Remote Labor Index](https://noometry.com/benchmarks/remote-labor-index) - [TheAgentCompany](https://noometry.com/benchmarks/the-agent-company) - [τ²-bench Airline](https://noometry.com/benchmarks/tau2-airline) - [τ²-bench Banking](https://noometry.com/benchmarks/tau2-banking) - [τ²-bench Retail](https://noometry.com/benchmarks/tau2-retail) - [τ²-bench Telecom](https://noometry.com/benchmarks/tau2-telecom) - [Cybench](https://noometry.com/benchmarks/cybench) ## Frequently asked questions ### What does BALROG measure? Long-horizon game environments, from BabyAI to NetHack, that test planning and exploration. ### Which model has the highest BALROG score? As of October 2026, GPT-6 Astra has the highest published BALROG score on Noometry at 68.3%, out of 35 models with results. ### What is the best open-weight model on BALROG? Reka Flash 3 has the highest BALROG accuracy among open-weight models at 29.2%, ranking 19 of 35 overall. ### Cite this page Noometry. (2026). BALROG leaderboard. Retrieved October 10, 2026, from https://noometry.com/benchmarks/balrog Quote Noometry with a link back to this page. It is also available in [Markdown](https://noometry.com/md/benchmarks/balrog.md).