o3-mini vs o4-mini
Wins 0 of 7 areas
—
Wins 7 of 7 areas
Coding · Agents · Reasoning · Facts · Math · Long documents · Following instructions
o4-mini is the stronger all-rounder.
Scores updated · 50 tests both models report · How we compare
Where each one wins
Tests won in each of the seven 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.
- ReasoningHard problems that need careful thinking03o4-mini3 of 4 tests · 1 tie
- CodingWriting and fixing software03o4-mini3 of 3 tests
- FactsGetting facts right instead of making them up02o4-mini2 of 3 tests · 1 tie
- MathCompetition and research-level math02o4-mini2 of 2 tests
- Following instructionsDoing exactly what it is asked02o4-mini2 of 2 tests
- AgentsCarrying out multi-step tasks on its own01o4-mini1 of 1 test
- Long documentsFinding answers in very long texts01o4-mini1 of 1 test
Agents and long documents rest on a single test each.
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.
The biggest differences
The tests each model wins by the widest margin, up to three each. Scores are out of 100.
Where o4-mini pulls ahead
- Fixes real GitHub issues in Python projectsSWE-bench Verified+18.8points ahead
- Reasons across sets of long documentsAA-LCR+18points ahead
- Unpublished advanced math problemsFrontierMath Tiers 1-3 (v2)+17.5points ahead
Where o3-mini pulls ahead
No clear win on a test scored out of 100.
Every test, side by side
All 50 tests both models report. The winning score is in its model's colour; marks a score checked independently.
Codingo4-mini
- SWE-bench Verifiedo4-mini by 18.849.368.1+18.8
- Terminal-Bench Hardo4-mini by 9.16.115.2+9.1
- SciCodeo4-mini by 3.742.846.5+3.7
Reasoningo4-mini
- Humanity's Last Examo4-mini by 4.51216.5+4.5
- ARC-AGI-2o4-mini by 3.136.1+3.1
- GPQA Diamondo4-mini by 1.177.378.4+1.1
- CritPttie0.30.6tie
Factso4-mini
- AA-Omniscience · Accuracyo4-mini by 3.521.324.8+3.5
- SimpleQA Verifiedo4-mini by 3.515.318.8+3.5
- AA-Omniscience · Non-hallucinationtie18.919.5tie
Matho4-mini
- FrontierMath Tiers 1-3 (v2)o4-mini by 17.518.636.1+17.5
- FrontierMath Tier 4o4-mini by 4.904.9+4.9
Following instructionso4-mini
- Multi-Challengeo4-mini by 3.139.943+3.1
- IFBencho4-mini by 1.667.168.7+1.6
Other results34 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.
- AIME 2025o4-mini by 44.448.392.7+44.4
- τ²-Bench Telecom (AA run)o4-mini by 24.331.355.6+24.3
- ARC-AGI-1o4-mini by 24.234.558.7+24.2
- usamo_2025o4-mini by 172.119.1+17
- TAU-bench (airline)o4-mini by 16.832.449.2+16.8
- TAU-bench (retail)o4-mini by 14.257.671.8+14.2
- Artificial Analysis Coding Indexo4-mini by 9.316.325.6+9.3
- MMLU Language (0-shot) - Swahilio4-mini by 7.573.881.3+7.5
- MMLU Language (0-shot) - Yorubao4-mini by 7.163.770.8+7.1
- AA-Omniscienceo4-mini by 6.9-42.6-35.7+6.9
- MMLU Language (0-shot) - Germano4-mini by 5.980.886.7+5.9
- AA Intelligenceo4-mini by 5.71116.7+5.7
- MMLU Language (0-shot) - Hindio4-mini by 4.881.185.9+4.8
- MMLU Language (0-shot) - Averageo4-mini by 4.580.785.2+4.5
- MMLU Language (0-shot) - Arabico4-mini by 4.281.986.1+4.2
- MMLU Language (0-shot) - Indonesiano4-mini by 4.182.886.9+4.1
- MMLU Language (0-shot) - Koreano4-mini by 4.182.686.7+4.1
- MMLU Language (0-shot) - Italiano4-mini by 3.983.887.7+3.9
- MMLU Language (0-shot) - Japaneseo4-mini by 3.883.186.9+3.8
- MMLU Language (0-shot) - Frencho4-mini by 3.783.787.4+3.7
- MMLU Language (0-shot) - Portuguese (Brazil)o4-mini by 3.784.187.8+3.7
- AIR-Bench 2024o4-mini by 3.674.978.5+3.6
- MMLU Language (0-shot) - Chinese (Simplified)o4-mini by 3.383.686.9+3.3
- XSTesto4-mini by 3.394.197.4+3.3
- Aider-Polygloto4-mini by 2.266.768.9+2.2
- frontiermath_tier_4_v1o4-mini by 2.14.26.3+2.1
- HarmBencho4-mini by 1.895.297+1.8
- livecodebench_hardo4-mini by 1.547.448.9+1.5
- livecodebench_mediumo3-mini by 1.387.586.2+1.3
- simple_safety_testso4-mini by 199100+1
- anthropic_red_teamtie98.698.2tie
- LiveCodeBenchtie77.777.4tie
- livecodebench_easytie99.198.8tie
- bbqtie94.194tie
Questions people ask
Which is better, o3-mini or o4-mini?
o4-mini wins all seven areas where both have results: coding, agents, reasoning, facts, math, long documents and following instructions. o3-mini wins none.
Which is better for coding?
o4-mini. It wins 3 of the 3 coding tests both models report; o3-mini wins none.
Which is cheaper?
o3-mini costs $1.10 per million input tokens and $4.40 per million output tokens; o4-mini costs $1.10 and $4.40. For a million tokens read plus a million written, they cost about the same.
How do you compare the two?
We use the 50 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 34 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.