Gemini 2.5 Flash vs o3
Wins 0 of 7 areas
—
Wins 6 of 7 areas
Coding · Agents · Reasoning · Images and charts · Long documents · Following instructions
o3 is the stronger all-rounder.Gemini 2.5 Flash is cheaper.
Scores updated · 38 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 thinking04o34 of 5 tests · 1 tie
- CodingWriting and fixing software03o33 of 3 tests
- Images and chartsUnderstanding pictures, charts and video12o32 of 3 tests
- AgentsCarrying out multi-step tasks on its own01o31 of 1 test
- Long documentsFinding answers in very long texts01o31 of 1 test
- Following instructionsDoing exactly what it is asked01o31 of 1 test
- FactsGetting facts right instead of making them up11Even1 each
Agents, long documents and following instructions 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.
Gemini 2.5 Flash costs 72% 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
- Hard command-line tasks in a real terminalTerminal-Bench Hard+23.5points ahead
- Follows unfamiliar, precisely checkable instructionsIFBench+21.1points ahead
- Answers hard knowledge questions correctlyAA-Omniscience · Accuracy+12.6points ahead
Where Gemini 2.5 Flash pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+12.8points ahead
Every test, side by side
All 38 tests both models report. The winning score is in its model's colour; marks a score checked independently.
Codingo3
- Terminal-Bench Hardo3 by 23.513.637.1+23.5
- SciCodeo3 by 11.929.141+11.9
- SWE-bench Verifiedo3 by 8.760.469.1+8.7
Reasoningo3
- SimpleBencho3 by 11.941.253.1+11.9
- Humanity's Last Examo3 by 812.120.1+8
- ARC-AGI-2o3 by 42.56.5+4
- GPQA Diamondo3 by 3.77982.7+3.7
- CritPttie1.11.1tie
FactsEven
- AA-Omniscience · Non-hallucinationGemini 2.5 Flash by 12.824.711.9+12.8
- AA-Omniscience · Accuracyo3 by 12.625.938.6+12.6
Images and chartso3
- MMMUo3 by 3.279.782.9+3.2
- LMArena · VisionGemini 2.5 Flash by 21 rating points12351214+21 rating
- MMMU-Proo3 by 169.170.1+1
Other results22 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 49.131.680.7+49.1
- HarmBencho3 by 35.862.698.4+35.8
- ARC-AGI-1o3 by 28.532.360.8+28.5
- SimpleQAo3 by 22.526.949.4+22.5
- Aider-Polygloto3 by 19.461.981.3+19.4
- livecodebench_hardo3 by 17.748.366+17.7
- AA Agentic Indexo3 by 17.318.836.1+17.3
- Artificial Analysis Coding Indexo3 by 16.222.238.4+16.2
- AIR-Bench 2024o3 by 15.868.784.5+15.8
- AIME 2025o3 by 14.47286.4+14.4
- AA-Omniscienceo3 by 14.3-29.8-15.6+14.3
- HMMT Feb. 2025o3 by 13.364.277.5+13.3
- AIME 2024o3 by 9.382.391.6+9.3
- LiveCodeBencho3 by 8.576.284.7+8.5
- livecodebench_mediumo3 by 7.382.589.8+7.3
- AA Intelligenceo3 by 7.113.120.2+7.1
- frontiermath_tier_4_v1Gemini 2.5 Flash by 2.14.22.1+2.1
- XSTestGemini 2.5 Flash by 1.598.897.3+1.5
- simple_safety_testso3 by 19899+1
- anthropic_red_teamtie98.898.3tie
- bbqtie97.797.9tie
- livecodebench_easytie99.199.1tie
Questions people ask
Which is better, Gemini 2.5 Flash or o3?
o3 wins six of the seven areas where both have results: coding, agents, reasoning, images and charts, long documents and following instructions. Gemini 2.5 Flash wins none, but costs 72% less. They are level on facts.
Which is better for coding?
o3. It wins 3 of the 3 coding tests both models report; Gemini 2.5 Flash wins none.
Which is cheaper?
Gemini 2.5 Flash costs $0.30 per million input tokens and $2.50 per million output tokens; o3 costs $2.00 and $8.00. That makes Gemini 2.5 Flash about 72% cheaper for the same work.
How do you compare the two?
We use the 38 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 22 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.