GPT-4o vs o3-mini
Wins 2 of 6 areas
Facts · Long documents
Wins 3 of 6 areas
Coding · Reasoning · Following instructions
o3-mini wins more areas, narrowly.GPT-4o is better at facts.
Scores updated · 37 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.
- ReasoningHard problems that need careful thinking03o3-mini3 of 4 tests · 1 tie
- CodingWriting and fixing software12o3-mini2 of 3 tests
- Following instructionsDoing exactly what it is asked01o3-mini1 of 2 tests · 1 tie
- FactsGetting facts right instead of making them up30GPT-4o3 of 3 tests
- Long documentsFinding answers in very long texts10GPT-4o1 of 1 test
- AgentsCarrying out multi-step tasks on its own00Even0 each · 1 tie
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.
o3-mini costs 73% less for the same work.
The biggest differences
The three tests each model wins by the widest margin. Scores are out of 100.
Where o3-mini pulls ahead
- Follows unfamiliar, precisely checkable instructionsIFBench+31.1points ahead
- Graduate-level biology, physics and chemistry questionsGPQA Diamond+24.7points ahead
- Fixes real GitHub issues in Python projectsSWE-bench Verified+16.1points ahead
Where GPT-4o pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+43.2points ahead
- Reasons across sets of long documentsAA-LCR+13points ahead
- Short factual questions, answered correctlySimpleQA Verified+10.7points ahead
Every test, side by side
All 37 tests both models report. The winning score is in its model's colour; marks a score checked independently.
Codingo3-mini
- SWE-bench Verifiedo3-mini by 16.133.249.3+16.1
- SciCodeo3-mini by 9.533.342.8+9.5
- Terminal-Bench HardGPT-4o by 2.28.36.1+2.2
Reasoningo3-mini
- GPQA Diamondo3-mini by 24.752.677.3+24.7
- Humanity's Last Examo3-mini by 10.21.812+10.2
- ARC-AGI-2o3-mini by 303+3
- CritPttie00.3tie
FactsGPT-4o
- AA-Omniscience · Non-hallucinationGPT-4o by 43.262.118.9+43.2
- SimpleQA VerifiedGPT-4o by 10.72615.3+10.7
- AA-Omniscience · AccuracyGPT-4o by 2.423.721.3+2.4
Following instructionso3-mini
- IFBencho3-mini by 31.13667.1+31.1
- Multi-Challengetie40.339.9tie
Other results23 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 2024o3-mini by 70.39.379.6+70.3
- livecodebench_mediumo3-mini by 55.432.187.5+55.4
- livecodebench_hardo3-mini by 42.94.547.4+42.9
- LiveCodeBencho3-mini by 39.438.377.7+39.4
- AIME 2025o3-mini by 36.611.748.3+36.6
- Aider-Polygloto3-mini by 3630.766.7+36
- AA-OmniscienceGPT-4o by 32.1-10.5-42.6+32.1
- ARC-AGI-1o3-mini by 304.534.5+30
- SimpleQAGPT-4o by 23.238.215+23.2
- livecodebench_easyo3-mini by 16.682.599.1+16.6
- IFEvalo3-mini by 12.98193.9+12.9
- AIR-Bench 2024o3-mini by 12.562.474.9+12.5
- HarmBencho3-mini by 12.382.995.2+12.3
- TAU-bench (airline)GPT-4o by 10.442.832.4+10.4
- Artificial Analysis Coding IndexGPT-4o by 7.924.216.3+7.9
- AA Intelligenceo3-mini by 3.77.311+3.7
- XSTestGPT-4o by 3.297.394.1+3.2
- TAU-bench (retail)GPT-4o by 2.760.357.6+2.7
- τ²-Bench Telecom (AA run)o3-mini by 2.428.931.3+2.4
- MGSMo3-mini by 1.590.592+1.5
- bbqGPT-4o by 195.194.1+1
- anthropic_red_teamtie99.198.6tie
- simple_safety_teststie98.599tie
Questions people ask
Which is better, GPT-4o or o3-mini?
o3-mini wins three of the six areas where both have results: coding, reasoning and following instructions. GPT-4o wins facts and long documents. They are level on agents.
Which is better for coding?
o3-mini. It wins 2 of the 3 coding tests both models report; GPT-4o wins 1.
Which is cheaper?
GPT-4o costs $5.00 per million input tokens and $15.00 per million output tokens; o3-mini costs $1.10 and $4.40. That makes o3-mini about 73% cheaper for the same work.
How do you compare the two?
We use the 37 benchmark tests both models have published scores on. The verdict counts the 14 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 23 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.