Claude 3.7 Sonnet vs o3
Wins 1 of 7 areas
Agents
Wins 4 of 7 areas
Reasoning · Images and charts · Long documents · Following instructions
o3 wins more areas, narrowly.Claude 3.7 Sonnet is better at agents.
Scores updated · 36 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 thinking03o33 of 4 tests · 1 tie
- Images and chartsUnderstanding pictures, charts and video03o33 of 3 tests
- Following instructionsDoing exactly what it is asked02o32 of 2 tests
- Long documentsFinding answers in very long texts01o31 of 1 test
- AgentsCarrying out multi-step tasks on its own10Claude 3.7 Sonnet1 of 1 test
- CodingWriting and fixing software11Even1 each · 1 tie
- FactsGetting facts right instead of making them up11Even1 each
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 costs 44% 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 pulls ahead
- Follows unfamiliar, precisely checkable instructionsIFBench+23.1points ahead
- Reasons across sets of long documentsAA-LCR+23points ahead
- Hard command-line tasks in a real terminalTerminal-Bench Hard+15.9points ahead
Where Claude 3.7 Sonnet pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+48.1points ahead
- Real work tasks from 44 professionsGDPVal+14.6points ahead
- Fixes real GitHub issues in Python projectsSWE-bench Verified+1.2points ahead
Every test, side by side
All 36 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingEven
- Terminal-Bench Hardo3 by 15.921.237.1+15.9
- SWE-bench VerifiedClaude 3.7 Sonnet by 1.270.369.1+1.2
- SciCodetie40.341tie
Reasoningo3
- Humanity's Last Examo3 by 10.49.720.1+10.4
- SimpleBencho3 by 6.746.453.1+6.7
- GPQA Diamondo3 by 5.577.282.7+5.5
- CritPttie0.91.1tie
FactsEven
- AA-Omniscience · Non-hallucinationClaude 3.7 Sonnet by 48.16011.9+48.1
- AA-Omniscience · Accuracyo3 by 10.62838.6+10.6
Images and chartso3
- MMMU-Proo3 by 1060.170.1+10
- MMMUo3 by 7.97582.9+7.9
- LMArena · Visiono3 by 45 rating points11691214+45 rating
Following instructionso3
- IFBencho3 by 23.148.371.4+23.1
- Multi-Challengeo3 by 8.851.660.4+8.8
Other results20 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.
- HMMT Feb. 2025o3 by 45.831.777.5+45.8
- ARC-AGI-1o3 by 32.228.660.8+32.2
- AIME 2025o3 by 31.654.886.4+31.6
- τ²-Bench Telecom (AA run)o3 by 2654.780.7+26
- Aider-Polygloto3 by 20.960.481.3+20.9
- metr_re_bencho3 by 15015+15
- AA-OmniscienceClaude 3.7 Sonnet by 14.8-0.7-15.6+14.8
- HarmBencho3 by 14.184.398.4+14.1
- metr_hcasto3 by 10.554.264.7+10.5
- TAU-bench (retail)Claude 3.7 Sonnet by 7.381.273.9+7.3
- TAU-bench (airline)Claude 3.7 Sonnet by 6.458.452+6.4
- bbqo3 by 5.892.197.9+5.8
- AIR-Bench 2024o3 by 2.781.884.5+2.7
- AA Intelligenceo3 by 2.517.720.2+2.5
- Artificial Analysis Coding Indexo3 by 236.438.4+2
- MATH-500 (EM)o3 by 1.996.298.1+1.9
- anthropic_red_teamClaude 3.7 Sonnet by 1.499.798.3+1.4
- simple_safety_testsClaude 3.7 Sonnet by 110099+1
- AA Agentic Indextie3736.1tie
- XSTesttie96.497.3tie
Questions people ask
Which is better, Claude 3.7 Sonnet or o3?
o3 wins four of the seven areas where both have results: reasoning, images and charts, long documents and following instructions. Claude 3.7 Sonnet wins agents. They are level on coding and facts.
Which is better for coding?
Neither. They win 1 coding test each of the 3 both models report, and 1 is a tie.
Which is cheaper?
Claude 3.7 Sonnet costs $3.00 per million input tokens and $15.00 per million output tokens; o3 costs $2.00 and $8.00. That makes o3 about 44% cheaper for the same work.
How do you compare the two?
We use the 36 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 20 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.