Side-by-side benchmark comparison across agentic, coding, multimodal, knowledge, reasoning, and math workflows.
GPT-OSS 20B is clearly ahead on the aggregate, 35 to 31. The gap is large enough that you do not need to squint at the spreadsheet to see the difference.
GPT-OSS 20B's sharpest advantage is in coding, where it averages 14.5 against 7.2. The single biggest benchmark swing on the page is SWE-bench Pro, 18 to 6. Ministral 3 3B (Reasoning) does hit back in instruction following, so the answer changes if that is the part of the workload you care about most.
Ministral 3 3B (Reasoning) is the reasoning model in the pair, while GPT-OSS 20B is not. That usually helps on harder chain-of-thought-heavy tests, but it can also mean more latency and more token spend in real use.
Pick GPT-OSS 20B if you want the stronger benchmark profile. Ministral 3 3B (Reasoning) only becomes the better choice if instruction following is the priority or you want the stronger reasoning-first profile.
GPT-OSS 20B
35.4
Ministral 3 3B (Reasoning)
34
GPT-OSS 20B
14.5
Ministral 3 3B (Reasoning)
7.2
GPT-OSS 20B
36
Ministral 3 3B (Reasoning)
30.4
GPT-OSS 20B
40.4
Ministral 3 3B (Reasoning)
35.3
GPT-OSS 20B
29
Ministral 3 3B (Reasoning)
25.2
GPT-OSS 20B
67
Ministral 3 3B (Reasoning)
68
GPT-OSS 20B
59.7
Ministral 3 3B (Reasoning)
59.7
GPT-OSS 20B
43.1
Ministral 3 3B (Reasoning)
40.9
GPT-OSS 20B is ahead overall, 35 to 31. The biggest single separator in this matchup is SWE-bench Pro, where the scores are 18 and 6.
GPT-OSS 20B has the edge for knowledge tasks in this comparison, averaging 29 versus 25.2. Inside this category, MMLU is the benchmark that creates the most daylight between them.
GPT-OSS 20B has the edge for coding in this comparison, averaging 14.5 versus 7.2. Inside this category, SWE-bench Pro is the benchmark that creates the most daylight between them.
GPT-OSS 20B has the edge for math in this comparison, averaging 43.1 versus 40.9. Inside this category, AIME 2023 is the benchmark that creates the most daylight between them.
GPT-OSS 20B has the edge for reasoning in this comparison, averaging 40.4 versus 35.3. Inside this category, LongBench v2 is the benchmark that creates the most daylight between them.
GPT-OSS 20B has the edge for agentic tasks in this comparison, averaging 35.4 versus 34. Inside this category, BrowseComp is the benchmark that creates the most daylight between them.
GPT-OSS 20B has the edge for multimodal and grounded tasks in this comparison, averaging 36 versus 30.4. Inside this category, MMMU-Pro is the benchmark that creates the most daylight between them.
Ministral 3 3B (Reasoning) has the edge for instruction following in this comparison, averaging 68 versus 67. Inside this category, IFEval is the benchmark that creates the most daylight between them.
GPT-OSS 20B and Ministral 3 3B (Reasoning) are effectively tied for multilingual tasks here, both landing at 59.7 on average.
Get notified when new models drop, benchmark scores change, or the leaderboard shifts. One email per week.
Free. No spam. Unsubscribe anytime. We only store derived location metadata for consent routing.