swe bench pro public Benchmark: Scores and Sources

Published result; benchmark version and evaluation conditions remain in the id and result note.

pillar: coding · weight 1 within pillar · unit: % (higher is better) · official board ↗

Row	Model	Result	Normalized (0–100)	Confidence
1	GPT-5.4 OpenAI	59.1%	59.1	100% confidence 100 percent, Full
2	Gemini 3.5 Flash Google	55.1%	55.1	100% confidence 100 percent, Full
3	Claude Opus 4.6 Anthropic	51.9%	51.9	100% confidence 100 percent, Full
4	Muse Spark 1.1 Meta	51.5%	51.5	61% confidence 61 percent, Medium
5	Claude Opus 4.6 Anthropic	47.1%	47.1	100% confidence 100 percent, Full
6	Gemini 3.1 Pro Preview Google	46.1%	46.1	100% confidence 100 percent, Full
7	Claude Opus 4.5 Anthropic	45.9%	45.9	100% confidence 100 percent, Full
8	Claude Sonnet 4.5 (latest) Anthropic	43.6%	43.6	43% confidence 43 percent, Low
9	Claude Sonnet 4.5 (latest) Anthropic	43.6%	43.6	43% confidence 43 percent, Low
10	GPT-5.4 OpenAI	43.4%	43.4	100% confidence 100 percent, Full
11	Gemini 3 Pro Preview Google	43.3%	43.3	68% confidence 68 percent, Medium
12	Gemini 3 Pro Preview Google	43.3%	43.3	68% confidence 68 percent, Medium
13	Claude Sonnet 4 (latest) Anthropic	42.7%	42.7	6% confidence 6 percent, Low
14	GPT-5 OpenAI	41.8%	41.8	100% confidence 100 percent, Full
15	GPT-5 OpenAI	41.8%	41.8	100% confidence 100 percent, Full
16	GPT-5.2 Codex OpenAI	41.0%	41.0	43% confidence 43 percent, Low
17	GPT-5.2 Codex OpenAI	41.0%	41.0	43% confidence 43 percent, Low
18	Claude Haiku 4.5 (latest) Anthropic	39.5%	39.5	33% confidence 33 percent, Low
19	Qwen3-Coder 480B-A35B Instruct Alibaba / Qwen	38.7%	38.7	47% confidence 47 percent, Low
20	MiniMax-M2.1 minimax	36.8%	36.8	13% confidence 13 percent, Low
21	Gemini 3 Flash Preview Google	34.6%	34.6	93% confidence 93 percent, High
22	Gemini 3 Flash Preview Google	34.6%	34.6	93% confidence 93 percent, High
23	GPT-5.2 OpenAI	29.9%	29.9	100% confidence 100 percent, Full
24	GPT-5.2 OpenAI	29.9%	29.9	100% confidence 100 percent, Full
25	Claude Opus 4.5 Anthropic	23.4%	23.4	100% confidence 100 percent, Full
26	Qwen3 235B-A22B Alibaba / Qwen	21.4%	21.4	76% confidence 76 percent, Medium
27	Qwen3 235B-A22B Alibaba / Qwen	21.4%	21.4	76% confidence 76 percent, Medium
28	Claude Opus 4.1 (latest) Anthropic	17.8%	17.8	14% confidence 14 percent, Low
29	GPT OSS 120B OpenAI	16.2%	16.2	79% confidence 79 percent, Medium
30	Gemma 3 27B IT Google	11.4%	11.4	74% confidence 74 percent, Medium
31	GLM-4.6 Z.ai	9.7%	9.7	55% confidence 55 percent, Medium
32	GLM-4.6 Z.ai	9.7%	9.7	55% confidence 55 percent, Medium
33	Claude Sonnet 4 Anthropic	9.1%	9.1	100% confidence 100 percent, Full
34	Llama 4 Maverick 17B Instruct Meta	5.2%	5.2	78% confidence 78 percent, Medium
35	Llama 4 Maverick 17B Instruct Meta	5.2%	5.2	78% confidence 78 percent, Medium
36	GPT-4o OpenAI	3.6%	3.6	59% confidence 59 percent, Medium

Results are as published by the source behind each value (hover or tap the number). Benchmark scores are shown individually for every source/variant (a model may have multiple rows), and vary by version, harness and date; the normalized column uses the method’s fixed 0–100 scales and feeds the coding pillar of the SI Score.