GPT-5 mini vs gpt-oss-120b
Wins 6 of 6 areas
Coding · Agents · Reasoning · Facts · Long documents · Following instructions
Wins 0 of 6 areas
—
GPT-5 mini is the stronger all-rounder.gpt-oss-120b is cheaper.
Scores updated · 32 tests both models report · How we compare
Where each one wins
Tests won in each of the six areas where both have results. Each piece is one test, so a longer bar means more evidence; grey means the two scored within a point of each other.
- FactsGetting facts right instead of making them up40GPT-5 mini4 of 4 tests
- AgentsCarrying out multi-step tasks on its own20GPT-5 mini2 of 3 tests · 1 tie
- Following instructionsDoing exactly what it is asked20GPT-5 mini2 of 2 tests
- CodingWriting and fixing software21GPT-5 mini2 of 3 tests
- ReasoningHard problems that need careful thinking21GPT-5 mini2 of 3 tests
- Long documentsFinding answers in very long texts10GPT-5 mini1 of 1 test
Long documents rests on a single test.
What it costs
Prices per million tokens, roughly 750,000 words. The bars show the cost of a million tokens read plus a million written.
gpt-oss-120b costs 67% less for the same work.
The biggest differences
The tests each model wins by the widest margin, up to three each. Scores are out of 100.
Where GPT-5 mini pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+34.4points ahead
- Reasons across sets of long documentsAA-LCR+20.3points ahead
- Keeps track of context across a multi-turn chatMulti-Challenge+13.7points ahead
Where gpt-oss-120b pulls ahead
- Command-line tasks in a real terminalTerminal-Bench 2.1+22.5points ahead
Every test, side by side
All 32 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingGPT-5 mini
- Terminal-Bench 2.1gpt-oss-120b by 22.53.726.2+22.5
- Terminal-Bench HardGPT-5 mini by 9.833.323.5+9.8
- SciCodeGPT-5 mini by 53934+5
AgentsGPT-5 mini
- GDPValGPT-5 mini by 7.913.55.6+7.9
- τ-Bench V3 · BankingGPT-5 mini by 2.715.512.8+2.7
- Terminal-Bench 4.0tie00tie
ReasoningGPT-5 mini
- GPQA DiamondGPT-5 mini by 4.682.878.2+4.6
- Humanity's Last ExamGPT-5 mini by 1.921.519.6+1.9
- CritPtgpt-oss-120b by 1.101.1+1.1
FactsGPT-5 mini
- AA-Omniscience · Non-hallucinationGPT-5 mini by 34.443.69.2+34.4
- SimpleQA VerifiedGPT-5 mini by 7.721.613.9+7.7
- AA-Omniscience · AccuracyGPT-5 mini by 3.22521.8+3.2
- Vectara HHEM hallucination ratelower is betterGPT-5 mini by 1.312.914.2+1.3
Following instructionsGPT-5 mini
- Multi-ChallengeGPT-5 mini by 13.75945.3+13.7
- IFBenchGPT-5 mini by 6.475.469+6.4
Other results16 tests, not counted
Tests outside the eight areas. They are not counted above: several are summary scores built from other tests, or the same test under another name.
- AA-OmniscienceGPT-5 mini by 31.9-17.3-49.3+31.9
- Artificial Analysis Coding Indexgpt-oss-120b by 14.815.630.4+14.8
- HMMT Nov. 2025gpt-oss-120b by 5.884.290+5.8
- AA IntelligenceGPT-5 mini by 5.216.811.6+5.2
- XSTestGPT-5 mini by 597.792.7+5
- vectara_avg_summary_lengthGPT-5 mini by 34.5 rating points169.7135.2+34.5 rating
- HarmBenchgpt-oss-120b by 2.997.1100+2.9
- AA Agentic IndexGPT-5 mini by 2.78.96.2+2.7
- τ²-Bench Telecom (AA run)GPT-5 mini by 2.668.465.8+2.6
- AIR-Bench 2024gpt-oss-120b by 2.385.788+2.3
- bbqgpt-oss-120b by 2.296.398.5+2.2
- vectara_factual_consistencyGPT-5 mini by 1.387.185.8+1.3
- AIME 2025GPT-5 mini by 1.191.190+1.1
- anthropic_red_teamtie99.199.5tie
- simple_safety_teststie100100tie
- vectara_answer_ratetie99.999.9tie
Questions people ask
Which is better, GPT-5 mini or gpt-oss-120b?
GPT-5 mini wins all six areas where both have results: coding, agents, reasoning, facts, long documents and following instructions. gpt-oss-120b wins none, but costs 67% less.
Which is better for coding?
GPT-5 mini. It wins 2 of the 3 coding tests both models report; gpt-oss-120b wins 1.
Which is cheaper?
GPT-5 mini costs $0.25 per million input tokens and $2.00 per million output tokens; gpt-oss-120b costs $0.15 and $0.59. That makes gpt-oss-120b about 67% cheaper for the same work.
How do you compare the two?
We use the 32 benchmark tests both models have published scores on. The verdict counts the 16 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 16 are listed but not counted, because several are summary scores or repeat a test. Each score is the one shown on the model's own page.