GSM8k Benchmark Leaderboard

Grade School Math 8K, a dataset of 8.5K high-quality linguistically diverse grade school math word problems requiring multi-step reasoning and elementary arithmetic operations.

Leaderboard

Top 47 models on GSM8k Benchmark Leaderboard (scores from public evaluations).

Rank	Model	Score	Lab
1	Kimi K2 Instruct	97.3%	—
2	o1	97.1%	—
3	GPT-4.5	97.0%	—
4	Llama 3.1 405B Instruct	96.8%	—
5	Claude 3.5 Sonnet	96.4%	—
5	Claude 3.5 Sonnet	96.4%	—
7	Gemma 3 27B	95.9%	—
7	Qwen2.5 32B Instruct	95.9%	—
9	Qwen2.5 72B Instruct	95.8%	—
10	DeepSeek-V2.5	95.1%	—
11	Claude 3 Opus	95.0%	—
12	Nova Pro	94.8%	—
12	Qwen2.5 14B Instruct	94.8%	—
14	Nova Lite	94.5%	—
15	Gemma 3 12B	94.4%	—
16	Qwen3 235B A22B	94.4%	—
17	Mistral Large 2	93.0%	—
18	Claude 3 Sonnet	92.3%	—
18	Nova Micro	92.3%	—
20	Kimi K2 Base	92.1%	—
21	Qwen2.5 7B Instruct	91.6%	—
22	Llama 3.1 Nemotron 70B Instruct	91.4%	—
23	Qwen2.5-Coder 32B Instruct	91.1%	—
23	Qwen2 72B Instruct	91.1%	—
25	Gemini 1.5 Pro	90.8%	—
26	Grok-1.5	90.0%	—
27	Gemma 3 4B	89.2%	—
28	Claude 3 Haiku	88.9%	—
29	Phi-3.5-MoE-instruct	88.7%	—
29	Qwen2.5-Omni-7B	88.7%	—
31	Phi 4 Mini	88.6%	—
32	Jamba 1.5 Large	87.0%	—
33	Phi-3.5-mini-instruct	86.2%	—
33	Gemini 1.5 Flash	86.2%	—
35	Qwen2.5-Coder 7B Instruct	83.9%	—
36	Qwen2 7B Instruct	82.3%	—
37	Granite 3.3 8B Instruct	80.9%	—
38	Mistral Small 3 24B Base	80.7%	—
39	Llama 3.2 3B Instruct	77.7%	—
40	Jamba 1.5 Mini	75.8%	—
41	Gemma 2 27B	74.0%	—
42	Command R+	70.7%	—
43	IBM Granite 4.0 Tiny Preview	70.1%	—
44	Gemma 2 9B	68.6%	—
45	Gemma 3 1B	62.8%	—
46	Granite 3.3 8B Base	59.0%	—
47	ERNIE 4.5	25.2%	—

Models tracked

Models with gsm8k in their evaluation profile.

No models linked yet.

View task leaderboards →