o3 vs o3-mini
Wins 7 of 7 areas
Coding · Agents · Reasoning · Facts · Math · Long documents · Following instructions
Wins 0 of 7 areas
—
o3 is the stronger all-rounder.o3-mini is cheaper.
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 thinking40o34 of 5 tests · 1 tie
- Following instructionsDoing exactly what it is asked20o32 of 2 tests
- CodingWriting and fixing software21o32 of 3 tests
- FactsGetting facts right instead of making them up21o32 of 3 tests
- AgentsCarrying out multi-step tasks on its own10o31 of 1 test
- MathCompetition and research-level math10o31 of 1 test
- Long documentsFinding answers in very long texts10o31 of 1 test
Agents, math 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.
o3-mini costs 45% 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 o3 pulls ahead
- Short factual questions, answered correctlySimpleQA Verified+34.1points ahead
- Reasons across sets of long documentsAA-LCR+31.7points ahead
- Hard command-line tasks in a real terminalTerminal-Bench Hard+31points ahead
Where o3-mini pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+7points ahead
- Code for real scientific research problemsSciCode+1.8points ahead
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.
Codingo3
- Terminal-Bench Hardo3 by 3137.16.1+31
- SWE-bench Verifiedo3 by 19.869.149.3+19.8
- SciCodeo3-mini by 1.84142.8+1.8
Reasoningo3
- SimpleBencho3 by 30.353.122.8+30.3
- Humanity's Last Examo3 by 8.120.112+8.1
- GPQA Diamondo3 by 5.482.777.3+5.4
- ARC-AGI-2o3 by 3.56.53+3.5
- CritPttie1.10.3tie
Factso3
- SimpleQA Verifiedo3 by 34.149.415.3+34.1
- AA-Omniscience · Accuracyo3 by 17.338.621.3+17.3
- AA-Omniscience · Non-hallucinationo3-mini by 711.918.9+7
Following instructionso3
- Multi-Challengeo3 by 20.560.439.9+20.5
- IFBencho3 by 4.371.467.1+4.3
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.
- τ²-Bench Telecom (AA run)o3 by 49.480.731.3+49.4
- HMMT Feb. 2025o3 by 49.277.528.3+49.2
- AIME 2025o3 by 38.186.448.3+38.1
- SimpleQAo3 by 34.449.415+34.4
- AA-Omniscienceo3 by 27.1-15.6-42.6+27.1
- ARC-AGI-1o3 by 26.360.834.5+26.3
- Artificial Analysis Coding Indexo3 by 22.138.416.3+22.1
- TAU-bench (airline)o3 by 19.65232.4+19.6
- livecodebench_hardo3 by 18.66647.4+18.6
- TAU-bench (retail)o3 by 16.373.957.6+16.3
- Aider-Polygloto3 by 14.681.366.7+14.6
- AIME 2024o3 by 1291.679.6+12
- MMLU Language (0-shot) - Germano3 by 9.790.580.8+9.7
- AIR-Bench 2024o3 by 9.684.574.9+9.6
- AA Intelligenceo3 by 9.220.211+9.2
- MMLU Language (0-shot) - Hindio3 by 8.789.881.1+8.7
- MMLU Language (0-shot) - Arabico3 by 8.590.481.9+8.5
- MMLU Language (0-shot) - Averageo3 by 8.188.880.7+8.1
- MMLU Language (0-shot) - Bengalio3 by 7.787.880.1+7.7
- MMLU Language (0-shot) - Italiano3 by 7.491.283.8+7.4
- LiveCodeBencho3 by 784.777.7+7
- MMLU Language (0-shot) - Indonesiano3 by 789.882.8+7
- MMLU Language (0-shot) - Frencho3 by 6.990.683.7+6.9
- MMLU Language (0-shot) - Koreano3 by 6.789.382.6+6.7
- FrontierMath (overall)o3 by 6.615.89.2+6.6
- MMLU Language (0-shot) - Chinese (Simplified)o3 by 5.789.383.6+5.7
- bbqo3 by 3.897.994.1+3.8
- HarmBencho3 by 3.298.495.2+3.2
- XSTesto3 by 3.297.394.1+3.2
- livecodebench_mediumo3 by 2.389.887.5+2.3
- frontiermath_tier_4_v1o3-mini by 2.12.14.2+2.1
- anthropic_red_teamtie98.398.6tie
- livecodebench_easytie99.199.1tie
- simple_safety_teststie9999tie
Questions people ask
Which is better, o3 or o3-mini?
o3 wins all seven areas where both have results: coding, agents, reasoning, facts, math, long documents and following instructions. o3-mini wins none, but costs 45% less.
Which is better for coding?
o3. It wins 2 of the 3 coding tests both models report; o3-mini wins 1.
Which is cheaper?
o3 costs $2.00 per million input tokens and $8.00 per million output tokens; o3-mini costs $1.10 and $4.40. That makes o3-mini about 45% cheaper for the same work.
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.