Vision & documents

SWE-bench Multimodal leaderboard

Name: benchgap: LLM benchmark scores, measured and estimated
Creator: benchgap
License: https://creativecommons.org/licenses/by-nc/4.0/

As of 2026-10-29, the highest measured score on SWE-bench Multimodal is 61.4% by Claude Opus 5.5. 31 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#	Model	Score	Source
1	Claude Opus 5.5	61.4%	measured
2	Claude Opus 5	59.4%	measured
3	Gemini 3 Flash	55.8%	estimated ± 7.2 pp, low confidence
4	GPT-5.6 Sol	55.1%	estimated ± 7.2 pp, low confidence
5	Claude Mythos 5	54.9%	measured
6	Claude Fable 5.1	54.7%	measured
7	Claude Sonnet 5.5	54.3%	measured
8	GPT-6.1 Terra	54.1%	estimated ± 7.2 pp, low confidence
9	Gemini 3.5 Flash	53.8%	estimated ± 7.2 pp, low confidence
10	Gemini 3.8 Flash	51.3%	estimated ± 7.2 pp, low confidence
11	GLM-5.3-Flash	50.2%	estimated ± 7.2 pp, low confidence
12	Gemini 3.7 Flash	50.1%	estimated ± 7.2 pp, low confidence
13	GPT-6 Astra	48.9%	estimated ± 7.2 pp, low confidence
14	Claude Fable 5	48.8%	estimated ± 7.2 pp, low confidence
15	Qwen3.8-27B	48.7%	estimated ± 7.2 pp, low confidence
16	GPT-6.1 Sol	48.7%	estimated ± 7.2 pp, low confidence
17	Gemini 3.6 Flash	48.2%	estimated ± 7.2 pp, low confidence
18	Gemini 3.1 Pro	47.8%	estimated ± 7.2 pp, low confidence
19	GPT-6 Sol	47.4%	estimated ± 7.2 pp, low confidence
20	GPT-5.6 Terra	44.6%	estimated ± 7.2 pp, low confidence
21	GPT-5.6 Luna	44.2%	estimated ± 7.2 pp, low confidence
22	GPT-5.5	43.8%	estimated ± 7.2 pp, low confidence
23	DeepSeek V4.1 Flash 1017	40.3%	estimated ± 7.2 pp, low confidence
24	GPT-6 Luna	39.7%	estimated ± 7.2 pp, low confidence
25	Claude Opus 4.8	38.4%	measured
26	Claude Opus 4.7	38.1%	estimated ± 7.2 pp, low confidence
27	Gemma 4 31B	36.4%	estimated ± 7.2 pp, low confidence
28	GPT-5.4	36.1%	estimated ± 7.2 pp, low confidence
29	Gemini 3.5 Flash-Lite	31.4%	estimated ± 7.2 pp, low confidence
30	Gemma 4 26B A4B	29.2%	estimated ± 7.2 pp, low confidence
31	Gemini 3.1 Flash-Lite	28.9%	estimated ± 7.2 pp, low confidence
32	Claude Sonnet 5	28.1%	measured
33	Claude Haiku 5.5	24.9%	estimated ± 7.2 pp, low confidence
34	Claude Opus 4.6	21.3%	estimated ± 7.2 pp, low confidence
35	Claude Haiku 4.5 Thinking	19.7%	estimated ± 7.2 pp, low confidence
36	Nemotron 3 Nano Omni 30B A3B	12.8%	estimated ± 7.2 pp, low confidence
37	Claude Haiku 4.5	8.9%	estimated ± 7.2 pp, low confidence
38	Qwen3.6-35B-A3B	7.8%	estimated ± 7.2 pp, low confidence

SWE-bench Multimodal leaderboard

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General