Skip to content
reasoning benchmark · included in ECI

DTBench

DTBench benchmark (score column: Accuracy).
Results
205
Random baseline
40.0%
Score ceiling
100%
Released
12 Aug 2026

Score by model release date

Best score by organisation

#ModelScoreRelativeEvidence
1Claude Opus 5.5
Anthropic
98.9%Source ↗
2Claude Fable 5
Anthropic
98.4%Source ↗
3Claude Opus 5
Anthropic
97.9%Source ↗
4Claude Opus 5 (xhigh)
Anthropic
97.9%Source ↗
5Claude Fable 5.1 (xhigh)
Anthropic
97.6%Source ↗
6Claude Fable 5.1 (unknown)
Anthropic
97.6%Source ↗
7GPT-6 Sol
OpenAI
97.3%Source ↗
8GPT-6 Astra (xhigh)
OpenAI
97.3%Source ↗
9Grok 4.6 (xhigh)
xAI
97.3%Source ↗
10Claude Fable 5 (xhigh)
Anthropic
97.3%Source ↗
11GPT-6 Astra (medium)
OpenAI
97.1%Source ↗
12GPT-6 Astra (unknown thinking)
OpenAI
97.1%Source ↗
13Claude Fable 5.1 (medium)
Anthropic
97.1%Source ↗
14Gemini 3.1 Pro Preview
Google
97.1%Source ↗
15Gemini 3.7 Flash
Google
96.8%Source ↗
16GPT-6 Astra
OpenAI
96.5%Source ↗
17Muse Spark 1.3 (unknown)
Meta AI
96.5%Source ↗
18Claude Fable 5.1
Anthropic
96.5%Source ↗
19Grok 4.5
SpaceXAI
96.5%Source ↗
20GPT-5.6 Sol (pro, max)
OpenAI
96.0%Source ↗
21GPT-5.5 Pro (xhigh)
OpenAI
96.0%Source ↗
22GPT-5.5 (xhigh)
OpenAI
96.0%Source ↗
23GPT-5.6 Sol (xhigh)
OpenAI
95.7%Source ↗
24Claude Fable 5 (medium)
Anthropic
95.7%Source ↗
25Gemini 3.8 Flash (unknown)
Google DeepMind
95.7%Source ↗
26Gemini 3.6 Flash
Google
95.5%Source ↗
27GPT-5.6 Sol
OpenAI
95.5%Source ↗
28GPT-5.6 Sol (medium)
OpenAI
95.2%Source ↗
29GPT-6 Astra (low)
OpenAI
94.9%Source ↗
30Gemini 3.8 Flash (medium)
Google DeepMind
94.9%Source ↗
31Gemini 3.8 Flash (low)
Google DeepMind
94.9%Source ↗
32Claude Opus 5 (medium)
Anthropic
94.9%Source ↗
33Claude Opus 4.8
Anthropic
94.9%Source ↗
34Gemini 3.7 Flash (medium)
Google DeepMind
94.7%Source ↗
35Muse Spark 1.2 (xhigh)
Meta AI
94.7%Source ↗
36Gemini 3.5 Flash
Google
94.7%Source ↗
37Claude Opus 4.7
Anthropic
94.7%Source ↗
38Muse Spark 1.1
Meta
94.4%Source ↗
39GPT-5.4 (xhigh)
OpenAI
94.4%Source ↗
40Claude Fable 5.1 (low)
Anthropic
94.1%Source ↗
41DeepSeek V4 Pro 0813 (unknown) Open source
DeepSeek
93.9%Source ↗
42GPT-5.6 Sol (low)
OpenAI
93.9%Source ↗
43GLM 5.2 Open source
Z.ai
93.6%Source ↗
44Gemini 3.7 Flash (low)
Google DeepMind
93.3%Source ↗
45GPT-5.6 Terra
OpenAI
93.3%Source ↗
46Claude Opus 5 (low)
Anthropic
93.1%Source ↗
47GPT-5.6 Terra (xhigh)
OpenAI
92.5%Source ↗
48Claude Sonnet 5
Anthropic
92.5%Source ↗
49Qwen3.7 Max
Qwen
92.3%Source ↗
50Qwen3.8 Max (xhigh)
Alibaba
92.0%Source ↗
51Claude Fable 5 (low)
Anthropic
91.5%Source ↗
52Kimi K3 Open source
MoonshotAI
91.2%Source ↗
53Claude Opus 4.6
Anthropic
91.2%Source ↗
54DeepSeek V4 Flash 0731 Open source
DeepSeek
90.9%Source ↗
55Kimi K2.6 Open source
MoonshotAI
90.9%Source ↗
56GPT-5.2 (xhigh)
OpenAI
90.9%Source ↗
57DeepSeek v4 Pro (high) Open source
DeepSeek
90.7%Source ↗
58Grok 4.3 Beta
xAI
90.7%Source ↗
59GPT-5
OpenAI
90.7%Source ↗
60GPT-6 Luna
OpenAI
90.1%Source ↗
61GPT-5.6 Terra (low)
OpenAI
90.1%Source ↗
62GPT-5.6 Terra (medium)
OpenAI
90.1%Source ↗
63Nemotron 3 Ultra Open source
NVIDIA
90.1%Source ↗
64Grok 4.20
SpaceXAI
90.1%Source ↗
65GPT-5.1
OpenAI
90.1%Source ↗
66DeepSeek V4.1 Flash (unknown) Open source
DeepSeek
89.9%Source ↗
67Claude Sonnet 4.6
Anthropic
89.9%Source ↗
68Claude Opus 4.5
Anthropic
89.9%Source ↗
69Claude Sonnet 5 (xhigh)
Anthropic
89.5%Source ↗
70GPT-5.6 Luna (xhigh)
OpenAI
89.1%Source ↗
71Gemini 3 Flash Preview
Google
89.1%Source ↗
72GPT-5.6 Luna
OpenAI
88.8%Source ↗
73Qwen3.8 27B Open source
Qwen
88.0%Source ↗
74Grok 4.1 Fast
xAI
87.7%Source ↗
75DeepSeek V3.2 Exp Open source
DeepSeek
87.7%Source ↗
76GLM-5.3 (unknown thinking)
Z.ai (Zhipu AI)
87.7%Source ↗
77Inkling (xhigh) Open source
Thinking Machines
87.5%Source ↗
78Qwen3.5 397B A17B Open source
Qwen
87.5%Source ↗
79Qwen 3.6 Max (Preview)
Alibaba
87.2%Source ↗
80o3 Pro
OpenAI
86.9%Source ↗
81DeepSeek v4 Flash (max) Open source
DeepSeek
86.4%Source ↗
82DeepSeek-V3.2 (Thinking; Novita) Open source
DeepSeek
85.6%Source ↗
83o3
OpenAI
84.8%Source ↗
84MiMo-V2.5-Pro Open source
Xiaomi
84.5%Source ↗
85Qwen3.5-122B-A10B Open source
Qwen
84.3%Source ↗
86Qwen3.7 Plus
Qwen
84.0%Source ↗
87GPT-5.6 Sol (none)
OpenAI
83.7%Source ↗
88Gemini 3.5 Flash Lite
Google
83.5%Source ↗
89GPT-5.6 Luna (medium)
OpenAI
83.2%Source ↗
90Claude Sonnet 4.5 (unknown thinking)
Anthropic
83.2%Source ↗
91Qwen3.5-Flash
Qwen
82.9%Source ↗
92Gemma 4 31B Open source
Google
82.7%Source ↗
93Grok 4 Fast
xAI
82.7%Source ↗
94DeepSeek V3.1 Open source
DeepSeek
82.7%Source ↗
95Qwen3.5-27B Open source
Qwen
82.4%Source ↗
96Gemini 2.5 Pro (Jun 2025)
Google DeepMind
82.4%Source ↗
97Qwen3-Max-Instruct
Alibaba
82.1%Source ↗
98Qwen 3.6 Plus (2026-04-02)
Alibaba
81.9%Source ↗
99Claude Opus 4
Anthropic
81.6%Source ↗
100DeepSeek V3.1 Terminus Open source
DeepSeek
81.3%Source ↗
101Qwen Plus (Apr 2025)
Alibaba
81.1%Source ↗
102Qwen3.5 Plus 2026-02-15
Qwen
80.5%Source ↗
103GPT-5 Mini
OpenAI
80.5%Source ↗
104GPT-5.4 nano (xhigh)
OpenAI
80.3%Source ↗
105Qwen3-235B-A22B-Thinking (Jul 2025) Open source
Alibaba
80.3%Source ↗
106GPT-5.6 Luna (low)
OpenAI
80.0%Source ↗
107GPT-5.4 mini (xhigh)
OpenAI
80.0%Source ↗
108Qwen3.5-35B-A3B Open source
Qwen
80.0%Source ↗
109Claude Opus 4.1 (unknown thinking)
Anthropic
80.0%Source ↗
110Claude Sonnet 5 (medium)
Anthropic
79.2%Source ↗
111MiniMax M3 Open source
MiniMax
78.9%Source ↗
112GPT-5.6 Terra (none)
OpenAI
78.7%Source ↗
113Qwen3-235B-A22B-Instruct (Jul 2025) Open source
Alibaba
78.4%Source ↗
114Qwen3.6 27B Open source
Qwen
78.1%Source ↗
115o4 Mini
OpenAI
77.6%Source ↗
116Qwen 3.6 Flash (2026-04-16)
Alibaba
77.1%Source ↗
117Claude Sonnet 4 (unknown thinking)
Anthropic
77.1%Source ↗
118Gemini 3.1 Flash Lite
Google
76.8%Source ↗
119Gemini 2.5 Flash (Jun 2025)
Google DeepMind
76.5%Source ↗
120gpt-oss-120b Open source
OpenAI
76.3%Source ↗
121Qwen3 235B A22B Open source
Qwen
75.7%Source ↗
122Mistral Medium 3.5 Open source
Mistral
75.5%Source ↗
123Gemma 4 26B A4B Open source
Google
74.9%Source ↗
124o1
OpenAI
74.7%Source ↗
125Qwen3.6 35B A3B Open source
Qwen
73.9%Source ↗
126Claude Sonnet 5 (low)
Anthropic
73.6%Source ↗
127Claude Haiku 4.5 (unknown thinking)
Anthropic
73.6%Source ↗
128Qwen3.5-9B Open source
Qwen
71.2%Source ↗
129Mistral Small 4 Open source
Mistral
70.9%Source ↗
130Qwen3-30B-A3B-Thinking (Jul 2025) Open source
Alibaba
69.3%Source ↗
131GPT-5.6 Luna (none)
OpenAI
69.3%Source ↗
132GPT-4.1 Mini
OpenAI
68.8%Source ↗
133o3 Mini
OpenAI
68.8%Source ↗
134GPT-4.1
OpenAI
68.3%Source ↗
135gpt-oss-20b Open source
OpenAI
68.0%Source ↗
136Claude 3.5 Sonnet (Oct 2024)
Anthropic
67.8%Source ↗
137Claude 3.5 Sonnet (Jun 2024)
Anthropic
67.8%Source ↗
138Mistral Medium 3.1
Mistral
67.5%Source ↗
139Qwen3 32B Open source
Qwen
67.5%Source ↗
140Qwen3-30B-A3B-Instruct (Jul 2025) Open source
Alibaba
67.2%Source ↗
141Grok-2 (Dec 2024)
xAI
65.2%Source ↗
142Mistral Large 3 Open source
Mistral AI
65.1%Source ↗
143DeepSeek-V3 (Mar 2025) Open source
DeepSeek
64.8%Source ↗
144GPT-4o (May 2024)
OpenAI
64.5%Source ↗
145Qwen3 14B Open source
Qwen
64.0%Source ↗
146Gemini 2.0 Flash (Feb 2025)
Google DeepMind,Google
63.2%Source ↗
147Qwen2.5-72B Open source
Alibaba
62.9%Source ↗
148gemini-2.5-flash-lite-preview-06-17 (Default thinking length)
Google DeepMind
62.8%Source ↗
149DeepSeek V3.2 Open source
DeepSeek
62.7%Source ↗
150GPT-5 Nano
OpenAI
62.7%Source ↗
151GPT-4 (Jun 2023)
OpenAI
62.7%Source ↗
152Mistral Medium 3
Mistral
62.3%Source ↗
153Llama 4 Maverick Open source
Meta
61.9%Source ↗
154Claude 3 Opus
Anthropic
61.6%Source ↗
155GPT-4 Turbo (Nov 2023)
OpenAI
61.6%Source ↗
156Llama 3.1-405B Open source
Meta AI
61.4%Source ↗
157Cohere Command A Open source
Cohere
61.3%Source ↗
158Magistral Small 1.2 Open source
Mistral AI
61.3%Source ↗
159Mistral Large 2 (Jul 2024) Open source
Mistral AI
61.2%Source ↗
160Mistral Large 2 (Nov 2024) Open source
Mistral AI
60.8%Source ↗
161Qwen3 30B A3B Open source
Qwen
60.3%Source ↗
162Llama 3.1-70B Open source
Meta AI
60.0%Source ↗
163Mistral Small 3.2 Open source
Mistral AI
59.9%Source ↗
164Qwen3 8B Open source
Qwen
59.7%Source ↗
165Llama 3.3 70B Open source
Meta AI
59.5%Source ↗
166Gemini 1.5 Pro (Sept 2024)
Google DeepMind
59.0%Source ↗
167Mistral Small 3.1 Open source
Mistral AI
58.6%Source ↗
168Llama 4 Scout Open source
Meta
57.9%Source ↗
169Claude 3.5 Haiku (Oct 2024)
Anthropic
56.7%Source ↗
170Gemini 1.5 Pro (May 2024)
Google DeepMind
56.5%Source ↗
171Mistral Large
mistralai
55.9%Source ↗
172Mixtral 8x22B Open source
Mistral AI
55.1%Source ↗
173Command R+ Open source
Cohere,Cohere Labs (formerly Cohere for AI)
54.9%Source ↗
174GPT-4o-mini
OpenAI
54.4%Source ↗
175Llama 3-70B Open source
Meta AI
54.2%Source ↗
176Gemini 1.5 Flash (Sep 2024)
Google DeepMind
53.8%Source ↗
177Claude 3 Sonnet
Anthropic
53.6%Source ↗
178Mistral Medium
Mistral AI
53.3%Source ↗
179mistral-small-2402
53.0%Source ↗
180GPT-4.1 Nano
OpenAI
52.5%Source ↗
181Gemma 3 27B Open source
Google
52.5%Source ↗
182Gemini 2.0 Flash-Lite
Google DeepMind
52.5%Source ↗
183Claude 2
Anthropic
51.9%Source ↗
184Ministral 3B
Mistral AI
51.7%Source ↗
185Gemini 1.5 Flash (May 2024)
Google DeepMind
51.4%Source ↗
186Claude 2.1
Anthropic
51.0%Source ↗
187Gemma 3 4B Open source
Google
50.9%Source ↗
188Llama 3.1-8B Open source
Meta AI
50.9%Source ↗
189Claude 3 Haiku
Anthropic
50.1%Source ↗
190Gemini 1.5 Flash 8B
Google DeepMind
50.0%Source ↗
191Mixtral 8x7B Open source
Mistral AI
49.6%Source ↗
192Mistral Small v24.09 Open source
Mistral AI
49.5%Source ↗
193Gemma 3 12B Open source
Google
48.8%Source ↗
194Mistral Nemo Open source
Mistral
48.6%Source ↗
195GPT-3.5 Turbo (Jan 2024)
OpenAI
48.5%Source ↗
196Gemma 2 27B Open source
Google
48.0%Source ↗
197Qwen2.5-7B Open source
Alibaba
47.7%Source ↗
198c4ai-command-r-08-2024
46.4%Source ↗
199Gemini 1.0 Pro
Google DeepMind
45.9%Source ↗
200Claude Instant
Anthropic
45.8%Source ↗
201Ministral 8B Open source
Mistral AI
45.7%Source ↗
202Llama 3-8B Open source
Meta AI
43.9%Source ↗
203Mistral 7B v0.3 Open source
Mistral AI
42.5%Source ↗
204Llama 2-13B Open source
Meta AI
42.2%Source ↗
205Llama 2-70B Open source
Meta AI
41.6%Source ↗

Caveats: settings (reasoning effort, agent scaffold) differ between rows and materially affect scores; the best reported setting per model is shown. Source: Epoch AI, CC BY 4.0.