Agentic · tools

MCP-Tasks leaderboard

Name: benchgap: LLM benchmark scores, measured and estimated
Creator: benchgap

As of 2026-10-07, the highest measured score on MCP-Tasks is 74.2% by Qwen3.5 397B. 161 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#	Model	Score	Source
1	Claude 3 Haiku	100.0%	estimated ± 8.2 pp, low confidence
2	DeepSeek-R1	100.0%	estimated ± 8.2 pp, low confidence
3	DeepSeek V3	100.0%	estimated ± 8.2 pp, low confidence
4	DeepSeek V3.1	100.0%	estimated ± 8.2 pp, low confidence
5	DeepSeek V3.1 (Reasoning)	100.0%	estimated ± 8.2 pp, low confidence
6	Exaone 4.0 1.2B	100.0%	estimated ± 8.2 pp, low confidence
7	Exaone 4.0 32B	100.0%	estimated ± 8.2 pp, low confidence
8	Gemini 2.5 Flash	100.0%	estimated ± 8.2 pp, low confidence
9	Gemma 3 27B	100.0%	estimated ± 8.2 pp, low confidence
10	Gemma 4 12B	100.0%	estimated ± 8.2 pp, low confidence
11	Gemma 4 E2B	100.0%	estimated ± 8.2 pp, low confidence
12	Gemma 4 E4B	100.0%	estimated ± 8.2 pp, low confidence
13	GPT-4.1 nano	100.0%	estimated ± 8.2 pp, low confidence
14	GPT-4o	100.0%	estimated ± 8.2 pp, low confidence
15	Granite-4.0-350M	100.0%	estimated ± 8.2 pp, low confidence
16	Granite-4.0-H-1B	100.0%	estimated ± 8.2 pp, low confidence
17	Granite-4.0-H-350M	100.0%	estimated ± 8.2 pp, low confidence
18	K-EXAONE 2.0	100.0%	estimated ± 6.4 pp, low confidence
19	LFM2.5-VL-1.6B-Extract	100.0%	estimated ± 8.2 pp, low confidence
20	Llama 3.1 405B	100.0%	estimated ± 8.2 pp, low confidence
21	Llama 4 Maverick	100.0%	estimated ± 8.2 pp, low confidence
22	Llama 4 Scout	100.0%	estimated ± 8.2 pp, low confidence
23	Mercury 2.5	100.0%	estimated ± 3.6 pp, low confidence
24	Mistral Large 2	100.0%	estimated ± 8.2 pp, low confidence
25	Mistral Large 3	100.0%	estimated ± 8.2 pp, low confidence
26	Mistral Medium 3	100.0%	estimated ± 8.2 pp, low confidence
27	Mistral Medium 3.5 128B	100.0%	estimated ± 3.6 pp, low confidence
28	Mistral Small 4	100.0%	estimated ± 8.2 pp, low confidence
29	Mistral Small 4 (Reasoning)	100.0%	estimated ± 8.2 pp, low confidence
30	Nemotron 3 Nano 30B	100.0%	estimated ± 8.2 pp, low confidence
31	Nemotron Ultra 253B	100.0%	estimated ± 8.2 pp, low confidence
32	North Mini Code	100.0%	estimated ± 8.2 pp, low confidence
33	Nova Pro	100.0%	estimated ± 8.2 pp, low confidence
34	o3-mini	100.0%	estimated ± 8.2 pp, low confidence
35	Ornith-1.0-397B	100.0%	estimated ± 6.4 pp, low confidence
36	Phi-4	100.0%	estimated ± 8.2 pp, low confidence
37	Qwen3-Omni-30B-A3B-Instruct	100.0%	estimated ± 8.2 pp, low confidence
38	Qwen3-Omni-30B-A3B-Thinking	100.0%	estimated ± 8.2 pp, low confidence
39	Sarvam 30B	100.0%	estimated ± 8.2 pp, low confidence
40	Solar Pro 2	100.0%	estimated ± 8.2 pp, low confidence
41	Ultravox v0.6 Llama 3.3 70B	100.0%	estimated ± 8.2 pp, low confidence
42	Gemma 4 26B A4B	99.0%	estimated ± 8.2 pp, low confidence
43	Qwen3.6-27B	98.1%	estimated ± 6.4 pp, low confidence
44	Nemotron 3 Nano Omni 30B A3B	97.9%	estimated ± 8.2 pp, low confidence
45	GLM-4.5-Air	97.1%	estimated ± 8.2 pp, low confidence
46	Sarvam 105B	96.9%	estimated ± 8.2 pp, low confidence
47	DeepSeek V3 0324	96.8%	estimated ± 8.2 pp, low confidence
48	GPT-4.1	96.8%	estimated ± 8.2 pp, low confidence
49	Claude Opus 4.6	93.7%	estimated ± 6.4 pp, low confidence
50	Claude 4 Sonnet	93.5%	estimated ± 8.2 pp, low confidence
51	GPT-4.1 mini	93.1%	estimated ± 8.2 pp, low confidence
52	Ornith-1.0-35B	92.5%	estimated ± 6.4 pp, low confidence
53	Gemini 2.5 Pro	92.3%	estimated ± 8.2 pp, low confidence
54	Gemma 4 31B	88.6%	estimated ± 8.2 pp, low confidence
55	GPT-OSS 20B	88.4%	estimated ± 8.2 pp, low confidence
56	Claude Sonnet 4.6	88.3%	estimated ± 6.4 pp, low confidence
57	Kimi K2	87.9%	estimated ± 8.2 pp, low confidence
58	o1	86.9%	estimated ± 8.2 pp, low confidence
59	Step 3.7 Flash	86.9%	estimated ± 6.4 pp, low confidence
60	Grok 4.1 Fast	86.2%	estimated ± 8.2 pp, low confidence
61	GPT-OSS 120B	84.9%	estimated ± 8.2 pp, low confidence
62	Grok 4 Fast (Reasoning)	84.9%	estimated ± 8.2 pp, low confidence
63	Qwen3.8 Max	84.2%	estimated ± 6.0 pp, low confidence
64	Claude 4.1 Opus Thinking	81.3%	estimated ± 8.2 pp, low confidence
65	GPT-5.6 Sol	80.4%	estimated ± 7.0 pp, low confidence
66	Muse Spark	80.4%	estimated ± 6.4 pp, low confidence
67	Claude Opus 4.7	79.7%	estimated ± 8.2 pp, low confidence
68	MiMo-V2.5-Pro	79.7%	estimated ± 3.6 pp, low confidence
69	K-Exaone	79.5%	estimated ± 8.2 pp, low confidence
70	Qwen3 Max	79.5%	estimated ± 8.2 pp, low confidence
71	dots3-note Preview	79.1%	estimated ± 6.0 pp, low confidence
72	Grok 4	79.1%	estimated ± 8.2 pp, low confidence
73	Ornith-1.0-9B	79.0%	estimated ± 6.4 pp, low confidence
74	Grok Code Fast 1	78.6%	estimated ± 8.2 pp, low confidence
75	GPT-5.6 Luna	78.3%	estimated ± 7.0 pp, low confidence
76	GPT-5.6 Terra	78.2%	estimated ± 7.0 pp, low confidence
77	GLM-4.6	77.8%	estimated ± 8.2 pp, low confidence
78	MiMo-V2.5	77.5%	estimated ± 6.4 pp, low confidence
79	Muse Spark 1.1	77.1%	estimated ± 0.6 pp, low confidence
80	Claude Opus 5	77.1%	estimated ± 0.6 pp, low confidence
81	Step 5 Preview	77.1%	estimated ± 0.6 pp, low confidence
82	Kimi K3	77.1%	estimated ± 0.6 pp, low confidence
83	Hy4 preview	77.1%	estimated ± 0.6 pp, low confidence
84	Gemini 3.5 Flash	77.1%	estimated ± 0.6 pp, low confidence
85	Claude Opus 4.8	77.1%	estimated ± 0.6 pp, low confidence
86	Ornith-1.5-397B	77.1%	estimated ± 0.6 pp, low confidence
87	Inkling-Small	77.1%	estimated ± 0.6 pp, low confidence
88	Beam	77.1%	estimated ± 0.6 pp, low confidence
89	Claude Opus 4.7 (Adaptive)	77.0%	estimated ± 0.6 pp, low confidence
90	GLM-5.2	77.0%	estimated ± 0.6 pp, low confidence
91	Qwen3.7 Max	77.0%	estimated ± 0.6 pp, low confidence
92	Kimi K2.7 Code	77.0%	estimated ± 0.6 pp, low confidence
93	Muse Glimmer 30B	77.0%	estimated ± 0.6 pp, low confidence
94	Agents-A1-4B	77.0%	estimated ± 8.2 pp, low confidence
95	GPT-5.5	77.0%	estimated ± 0.6 pp, low confidence
96	MiniMax M3	77.0%	estimated ± 0.6 pp, low confidence
97	Inkling	77.0%	estimated ± 0.6 pp, low confidence
98	DeepSeek V4 Pro 0813	77.0%	estimated ± 0.6 pp, low confidence
99	Qwen3.7 Plus	77.0%	estimated ± 0.6 pp, low confidence
100	GLM-5.1	76.9%	estimated ± 0.6 pp, low confidence
101	GPT-5.4	76.9%	estimated ± 0.6 pp, low confidence
102	Ornith-1.5-35B-A3B	76.9%	estimated ± 0.6 pp, low confidence
103	DeepSeek V4 Flash 0731	76.9%	estimated ± 0.6 pp, low confidence
104	Ling 3.0 Flash	76.7%	estimated ± 0.6 pp, low confidence
105	Qwen3.6-35B-A3B	76.6%	estimated ± 0.6 pp, low confidence
106	Solar Pro 4	76.5%	estimated ± 0.6 pp, low confidence
107	Solar Open 2	76.3%	estimated ± 0.6 pp, low confidence
108	GPT-5.4 mini	76.2%	estimated ± 0.6 pp, low confidence
109	GPT-5.4 nano	76.0%	estimated ± 0.6 pp, low confidence
110	Kimi K2.6	76.0%	estimated ± 0.6 pp, low confidence
111	Ornith-1.5-9B	75.8%	estimated ± 0.6 pp, low confidence
112	Ternary Bonsai 2 27B	75.7%	estimated ± 8.2 pp, low confidence
113	o3	75.4%	estimated ± 8.2 pp, low confidence
114	Nemotron 3 Ultra	75.1%	estimated ± 3.6 pp, low confidence
115	GPT-5.1	74.7%	estimated ± 8.2 pp, low confidence
116	Qwen3.5 397B	74.2%	measured
117	Qwen3.6 Plus	74.1%	measured
118	GPT-5.1-Codex	74.0%	estimated ± 8.2 pp, low confidence
119	GPT-5.1-Codex-Max	74.0%	estimated ± 8.2 pp, low confidence
120	LLaDA2.2-flash	73.8%	estimated ± 0.6 pp, medium confidence
121	LongCat-Flash-Lite-Sparse	73.5%	estimated ± 0.6 pp, medium confidence
122	MiMo-V2-Flash	73.4%	estimated ± 8.2 pp, low confidence
123	Qwen3.5 397B (Reasoning)	73.4%	estimated ± 8.2 pp, low confidence
124	GPT-5.2	72.8%	estimated ± 8.2 pp, low confidence
125	GPT-5 (high)	72.8%	estimated ± 8.2 pp, low confidence
126	Command A+	72.7%	estimated ± 8.2 pp, low confidence
127	GPT-5.3 Codex	72.1%	estimated ± 8.2 pp, low confidence
128	Ling 2.6 Flash	72.1%	estimated ± 8.2 pp, low confidence
129	Solar Pro 3	71.9%	estimated ± 8.2 pp, low confidence
130	Claude Opus 4.5	71.8%	measured
131	GPT-5 (medium)	71.7%	estimated ± 8.2 pp, low confidence
132	Gemini 3 Pro	71.4%	estimated ± 8.2 pp, low confidence
133	LFM2.5-8B-A1B	70.7%	estimated ± 8.2 pp, low confidence
134	Qwen3.5-35B-A3B	70.0%	estimated ± 8.2 pp, low confidence
135	Claude Opus 4.5 Thinking	69.8%	estimated ± 8.2 pp, low confidence
136	Trinity-Large-Preview	69.5%	estimated ± 8.2 pp, low confidence
137	Trinity-Large-Thinking	69.5%	estimated ± 8.2 pp, low confidence
138	Gemini 3.1 Pro	69.3%	estimated ± 6.4 pp, low confidence
139	MiMo-V2-Pro	69.3%	estimated ± 6.4 pp, low confidence
140	Claude Opus 4.6 (Adaptive)	68.2%	estimated ± 8.2 pp, low confidence
141	GPT-5.2-Codex	68.2%	estimated ± 8.2 pp, low confidence
142	LLaDA2.2-mini	68.1%	estimated ± 6.4 pp, low confidence
143	Grok 4.1 Fast (Reasoning)	67.4%	estimated ± 8.2 pp, low confidence
144	Qwen3.5-122B-A10B	67.2%	estimated ± 8.2 pp, low confidence
145	Qwen3.5-27B	67.0%	estimated ± 8.2 pp, low confidence
146	GLM-4.7	65.8%	estimated ± 8.2 pp, low confidence
147	Kimi K2.5 (Reasoning)	65.8%	estimated ± 8.2 pp, low confidence
148	Qwen 3.6 Max (preview)	65.8%	estimated ± 8.2 pp, low confidence
149	GLM-5-Turbo	65.8%	estimated ± 6.4 pp, low confidence
150	Grok 4.3	64.6%	estimated ± 8.2 pp, low confidence
151	Claude Fable 5	64.1%	estimated ± 8.2 pp, low confidence
152	Pokee-Isaac 28B	63.1%	estimated ± 3.6 pp, medium confidence
153	GLM-5V-Turbo	62.4%	estimated ± 6.4 pp, low confidence
154	GLM-5	60.8%	measured
155	Kimi K2.5	59.1%	measured
156	Gemini 3 Flash	54.8%	estimated ± 6.4 pp, low confidence
157	MiniMax M2.7	54.0%	estimated ± 6.4 pp, low confidence
158	Granite 4.2 30B	51.3%	estimated ± 3.6 pp, low confidence
159	MiMo-V2-Omni	48.7%	estimated ± 6.4 pp, low confidence
160	DeepSeek V3.2	41.4%	estimated ± 6.4 pp, low confidence
161	Granite 4.2 8B	40.2%	estimated ± 3.6 pp, low confidence
162	Granite 4.2 3B	13.0%	estimated ± 3.6 pp, low confidence
163	Atria Dawn Preview	7.3%	estimated ± 3.6 pp, low confidence
164	Nemotron 3 Super 100B	0.8%	estimated ± 6.4 pp, low confidence
165	Nemotron 3.5 Lightning 30B A3B NVFP4	0.0%	estimated ± 3.6 pp, low confidence
166	LFM2.5-2.6B	0.0%	estimated ± 3.6 pp, low confidence

MCP-Tasks leaderboard

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General