All General Coding Reasoning Multimodal Safety Specialized Agentic Language Long Context Speed Text-to-Image Image Editing Text-to-Video Image-to-Video
AIME194 AIME 2025270 ARC-AGI Artificial Analysis Math Index270 BBH2839 Chatbot Arena Math274 FrontierMath GPQA (Open LLM Leaderboard)2839 GPQA Diamond615 GSM8K LogiQA MATH MATH Level 52839 MATH-500202 MuSR2839

Reasoning

Composite performance across 10 benchmarks · Normalized 0–100 · At least 5 scores required · Ranked by average performance · Top 50 of 172 eligible models

# Model Provider AIME AIME 2025 AA Math BBH Arena Math GPQA Open GPQA Dmd MATH L5 MATH-500 MuSR Avg Coverage Price Value
1 Grok 4 SpaceXAI 94.3% 92.7% 92.7 — 1,432 — 87.7% — 99.0% —
94.8
6/10 $18.00/M 5.3
2 GPT-5 (high) OpenAI 95.7% 94.3% 94.3 — 1,404 — 85.4% — 99.4% —
94.5
6/10 $11.25/M 8.4
3 GPT-5 (medium) OpenAI 91.7% 91.7% 91.7 — — — 84.2% — 99.1% —
93.3
5/10 $11.25/M 8.3
4 Qwen3 235B A22B 2507 (Reasoning) Alibaba 94.0% 91.0% 91.0 — — — 79.0% — 98.4% —
92.2
5/10 $0.64/M 144.1
5 OpenAI: o4 Mini High OpenAI 94.0% 90.7% 90.7 — — — 78.4% — 98.9% —
92.0
5/10 $5.50/M 16.7
6 Google: Gemini 2.5 ProRetiring Google 88.7% 87.7% 87.7 — 1,459 — 84.4% — 96.7% —
91.8
6/10 $11.25/M 8.2
7 OpenAI: o3 OpenAI 90.3% 88.3% 88.3 — — — 82.7% — 99.2% —
91.4
5/10 $10.00/M 9.1
8 Grok 3 mini Reasoning (high) SpaceXAI 93.3% 84.7% 84.7 — — — 79.1% — 99.2% —
89.7
5/10 $0.80/M 112.2
9 GPT-5 (low) OpenAI 83.0% 83.0% 83.0 — — — 80.8% — 98.7% —
87.2
5/10 $11.25/M 7.7
10 DeepSeek R1 0528 (May '25) DeepSeek 89.3% 76.0% 76.0 — — — 81.3% — 98.3% —
85.7
5/10 $4.35/M 19.7
11 EXAONE 4.0 32B (Reasoning) LG AI Research 84.3% 80.0% 80.0 — — — 73.9% — 97.7% —
84.4
5/10 — —
12 Llama Nemotron Super 49B v1.5 (Reasoning) NVIDIA 86.0% 76.7% 76.7 — — — 74.8% — 98.3% —
83.8
5/10 $0.80/M 104.7
13 GLM-4.5 (Reasoning) Z.ai 87.3% 73.7% 73.7 — — — 78.2% — 97.9% —
83.5
5/10 $2.80/M 29.8
14 Qwen3 235B A22B (Reasoning) Alibaba 84.0% 82.0% 82.0 — — — 70.0% — 93.0% —
83.2
5/10 $2.28/M 36.6
15 o1-preview OpenAI — 79.7% 79.7 — 1,376 — 76.5% — 92.4% —
83.0
5/10 $82.50/M 1.0
16 Gemini 2.5 Flash (Reasoning) Google 82.3% 73.3% 73.3 — — — 79.0% — 98.1% —
82.6
5/10 $2.80/M 29.5
17 Z.ai: GLM 4.5 Air Z.ai 67.3% 80.7% 80.7 — 1,403 — 73.3% — 96.5% —
82.1
6/10 $0.98/M 83.8
18 Claude 4 Sonnet (Reasoning) Anthropic 77.3% 74.3% 74.3 — — — 77.7% — 99.1% —
81.8
5/10 $18.00/M 4.5
19 Claude 4 Opus (Reasoning) Anthropic 75.7% 73.3% 73.3 — — — 79.6% — 98.2% —
81.3
5/10 $90.00/M 0.9
20 Qwen3 32B (Reasoning) Alibaba 80.7% 73.0% 73.0 — — — 66.8% — 96.1% —
78.8
5/10 $0.36/M 218.9
21 Qwen3 235B A22B 2507 Instruct Alibaba 71.7% 71.7% 71.7 — — — 75.3% — 98.0% —
78.8
5/10 $1.10/M 71.6
22 Qwen3 30B A3B (Reasoning) Alibaba 75.3% 72.3% 72.3 — — — 61.6% — 95.9% —
76.2
5/10 $0.62/M 122.9
23 MiniMax M1 80k MiniMax 84.7% 61.0% 61.0 — — — 69.7% — 98.0% —
75.9
5/10 $2.75/M 27.6
24 Qwen3 30B A3B 2507 (Reasoning) Alibaba 90.7% 56.3% 56.3 — — — 70.7% — 97.6% —
75.4
5/10 $1.66/M 45.4
25 DeepSeek R1 (Jan '25) DeepSeek 68.3% 68.0% 68.0 — — — 70.8% — 96.6% —
75.3
5/10 $6.00/M 12.5
26 Llama 3.1 Nemotron Ultra 253B v1 (Reasoning) NVIDIA 74.7% 63.7% 63.7 — — — 72.8% — 95.2% —
75.0
5/10 — —
27 Qwen3 30B A3B 2507 Instruct Alibaba 72.7% 66.3% 66.3 — — — 65.9% — 97.5% —
74.6
5/10 $0.60/M 124.3
28 Solar Pro 2 (Reasoning) Upstage 69.0% 61.3% 61.3 — — — 68.7% — 96.7% —
72.2
5/10 — —
29 DeepSeek R1 0528 Qwen3 8B DeepSeek 65.0% 63.7% 63.7 — — — 61.2% — 93.2% —
69.9
5/10 — —
30 Qwen3 14B (Reasoning) Alibaba 76.3% 55.7% 55.7 — — — 60.4% — 96.1% —
69.5
5/10 $0.36/M 192.9
31 Gemini 2.5 Flash-Lite (Reasoning) Google 70.3% 53.3% 53.3 — — — 62.5% — 96.9% —
67.9
5/10 $0.50/M 135.8
32 Claude 3.7 Sonnet (Reasoning) Anthropic 48.7% 56.3% 56.3 — — — 77.2% — 94.7% —
67.5
5/10 — —
33 Gemini 2.5 Flash (Non-reasoning) Google 50.0% 60.3% 60.3 — — — 68.3% — 93.2% —
67.0
5/10 $2.80/M 23.9
34 Llama 3.3 Nemotron Super 49B v1 (Reasoning) NVIDIA 58.3% 54.7% 54.7 — — — 64.3% — 95.9% —
66.1
5/10 — —
35 deepseek-ai/DeepSeek-R1-Distill-Qwen-14B DeepSeek 66.7% 55.7% 55.7 40.7% — 18.3% 48.4% 57.0% 94.9% 28.7%
65.7
9/10 — —
36 Qwen/Qwen2.5-72B-Instruct Alibaba — — — 61.9% 1,286 16.7% — 59.8% — 11.7%
64.4
5/10 — —
37 deepseek-v3-0324 DeepSeek 52.0% 41.0% 41.0 — 1,378 — 65.5% — 94.2% —
63.5
6/10 $1.14/M 55.7
38 Magistral Small 1 Mistral AI 71.3% 41.3% 41.3 — — — 64.1% — 96.3% —
63.5
5/10 — —
39 Magistral Medium 1 Mistral AI 70.0% 40.3% 40.3 — — — 67.9% — 91.7% —
62.7
5/10 — —
40 ERNIE 4.5 300B A47B Baidu 49.3% 41.3% 41.3 — — — 81.1% — 93.1% —
62.1
5/10 $1.38/M 45.0
41 Grok 3 SpaceXAI 33.0% 58.0% 58.0 — — — 69.3% — 87.0% —
61.4
5/10 $18.00/M 3.4
42 qwen3-coder-480b-a35b-instruct Alibaba 47.7% 39.3% 39.3 — 1,370 — 61.8% — 94.2% —
61.2
6/10 $2.10/M 29.2
43 Llama 3.1 Nemotron Nano 4B v1.1 (Reasoning) NVIDIA 70.7% 50.0% 50.0 — — — 40.8% — 94.7% —
61.1
5/10 — —
44 OpenAI: GPT-4.1 Mini OpenAI 43.0% 46.3% 46.3 — — — 66.4% — 92.5% —
59.3
5/10 $2.00/M 29.7
45 Claude 4 Opus (Non-reasoning) Anthropic 56.3% 36.3% 36.3 — — — 70.1% — 94.1% —
59.3
5/10 $90.00/M 0.7
46 Qwen/Qwen2-72B-Instruct Alibaba — — — 57.5% 1,238 16.3% — 41.8% — 17.2%
59.2
5/10 — —
47 allenai/Llama-3.1-Tulu-3-70B Meta — — — 45.3% 1,244 16.4% — 38.3% — 24.3%
59.0
5/10 — —
48 Meta: Llama 3.1 70B Instruct Meta — — — 55.9% 1,255 14.2% — 38.1% — 17.7%
57.1
5/10 $0.80/M 71.4
49 Meta: Llama 3.3 70B Instruct Meta — — — 56.6% 1,271 10.5% — 48.3% — 15.6%
57.1
5/10 $1.38/M 41.3
50 EXAONE 4.0 32B (Non-reasoning) LG AI Research 47.0% 39.3% 39.3 — — — 62.8% — 93.9% —
56.8
5/10 — —

Grok 4 leads on raw performance, but Qwen3 32B (Reasoning) delivers 4030% more performance per dollar. That extra 16.0 points costs $1.10/M per point.

Access this data

JSON GET /api/v1 
[
  {
    "rank": 1,
    "model": "Grok 4",
    "provider": "SpaceXAI",
    "avg_score": 94.8,
    "cost": 18.0,
    "cost_unit": "million_tokens",
    "value": 5.3
  },
  {
    "rank": 2,
    "model": "GPT-5 (high)",
    "provider": "OpenAI",
    "avg_score": 94.5,
    "cost": 11.25,
    "cost_unit": "million_tokens",
    "value": 8.4
  },
  {
    "rank": 3,
    "model": "GPT-5 (medium)",
    "provider": "OpenAI",
    "avg_score": 93.3,
    "cost": 11.25,
    "cost_unit": "million_tokens",
    "value": 8.3
  }
]
Curl 
$ curl -s https://rightrank.ai/api/v1/rankings/reasoning.json | jq '.[:3]'
CLI brew install rightrank 
$ rightrank top reasoning --limit 5
$ rightrank compare claude-opus-4.6 gpt-5.4-pro --category reasoning
$ rightrank value reasoning  # best performance per dollar
Claude Code Skill Skill 
# In Claude Code, use the RightRank skill:
/rightrank best model for reasoning
/rightrank compare top 5 by value
/rightrank cheapest model scoring above 80 on reasoning