GSM8K
AI benchmark results and model performance
Type: percentage
Recorded results
62
Average score
80.12%
Score range
25.20% - 97%
Leading model
97% - GPT 4.5 Preview
Scores over time
Individual benchmark scores plotted by date.
Model results
| Organisation | Models | Reported | Top score | Info | Self reported | Source |
|---|---|---|---|---|---|---|
| GPT 4.5 Preview | 27 Feb 2025 | 97% | - | Yes | Source | |
| Qwen 72B | - | 95.80% | inferred family alias from qwen-2.5-72b-instruct (score=0.3060; benches=14) | Yes | Source | |
| DeepSeek V2.5 (2024-09-05) | 05 Sept 2024 | 95.10% | inferred alias from deepseek-v2.5 | Yes | Source | |
| DeepSeek V2.5 (2024-12-10) | 10 Dec 2024 | 95.10% | inferred alias from deepseek-v2.5 | Yes | Source | |
| Qwen 14B | - | 94.80% | inferred family alias from qwen-2.5-14b-instruct (score=0.3060; benches=16) | Yes | Source | |
| Qwen 3 235B A22B | - | 94.39% | - | Yes | Source | |
| Granite 4.1 30B | 29 Apr 2026 | 94.16% | 8-shot | Yes | Source | |
| Mistral Large 2.1 | 18 Nov 2024 | 93% | inferred version-family alias from mistral-large-2-2407 | Yes | Source | |
| Mistral OCR 2 | 22 May 2025 | 93% | inferred family alias from mistral-large-2-2407 (score=0.3100; benches=5) | Yes | Source | |
| DeepSeek V4.1 Flash | 10 Sept 2026 | 93% | Base model; 8-shot; EM | Yes | Source | |
| DeepSeek V4 Pro | 24 Apr 2026 | 92.60% | Hugging Face evaluation result | Yes | Source | |
| Granite 4.1 8B | 29 Apr 2026 | 92.49% | 8-shot | Yes | Source | |
| Kimi K2 (2025-07-11) | 11 Jul 2025 | 92.10% | EM | Yes | Source | |
| Qwen 7B | - | 91.60% | inferred family alias from qwen-2.5-7b-instruct (score=0.3083; benches=14) | Yes | Source | |
| Llama 3.1 Nemotron 70B Instruct | 01 Oct 2024 | 91.43% | - | Yes | Source | |
| Qwen 2 Math 72B | - | 91.10% | inferred high-confidence family alias from qwen2-72b-instruct (score=0.4667; benches=17) | Yes | Source | |
| Qwen 2 Math RM 72B | - | 91.10% | inferred family alias from qwen2-72b-instruct (score=0.3917; benches=17) | Yes | Source | |
| Gemini 1.5 Pro 001 | 23 May 2024 | 90.80% | inferred alias from gemini-1.5-pro | Yes | Source | |
| Gemini 1.5 Pro Exp (2024-08-01) | 01 Aug 2024 | 90.80% | inferred alias from gemini-1.5-pro | Yes | Source | |
| Gemini 1.5 Pro Exp (2024-08-27) | 27 Aug 2024 | 90.80% | inferred alias from gemini-1.5-pro | Yes | Source | |
| Gemini Robotics ER 1.5 Preview | 25 Sept 2025 | 90.80% | inferred family alias from gemini-1.5-pro (score=0.3717; benches=23) | Yes | Source | |
| LearnLM 1.5 Pro Experimental | 19 Nov 2024 | 90.80% | inferred family alias from gemini-1.5-pro (score=0.3700; benches=23) | Yes | Source | |
| Grok 1.5 | 28 Mar 2024 | 90% | - | Yes | Source | |
| Phi 3.5 MoE instruct | 23 Aug 2024 | 88.70% | - | Yes | Source | |
| Qwen 2.5 Coder 3B | - | 88.70% | inferred family alias from qwen2.5-omni-7b (score=0.3000; benches=45) | Yes | Source |
