DeepSeek-R1 0528 vs o3
Wins 0 of 6 areas
—
Wins 6 of 6 areas
Coding · Agents · Reasoning · Facts · Long documents · Following instructions
o3 is the stronger all-rounder.DeepSeek-R1 0528 is cheaper.
Scores updated · 45 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 thinking03o33 of 4 tests · 1 tie
- CodingWriting and fixing software02o32 of 3 tests · 1 tie
- AgentsCarrying out multi-step tasks on its own02o32 of 2 tests
- Long documentsFinding answers in very long texts02o32 of 2 tests
- Following instructionsDoing exactly what it is asked02o32 of 2 tests
- FactsGetting facts right instead of making them up12o32 of 3 tests
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.
DeepSeek-R1 0528 costs 57% 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
- Finds hard-to-locate facts by browsing the webBrowseComp+40.8points ahead
- Follows unfamiliar, precisely checkable instructionsIFBench+31.8points ahead
- Fixes real GitHub issues in Python projectsSWE-bench Verified+24.5points ahead
Where DeepSeek-R1 0528 pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+4.7points ahead
Every test, side by side
All 45 tests both models report. The winning score is in its model's colour; marks a score checked independently.
Codingo3
- SWE-bench Verifiedo3 by 24.544.669.1+24.5
- Terminal-Bench Hardo3 by 21.215.937.1+21.2
- SciCodetie40.341tie
Reasoningo3
- SimpleBencho3 by 12.340.853.1+12.3
- Humanity's Last Examo3 by 4.315.820.1+4.3
- GPQA Diamondo3 by 1.481.382.7+1.4
- CritPttie1.41.1tie
Factso3
- SimpleQA Verifiedo3 by 2227.449.4+22
- AA-Omniscience · Accuracyo3 by 8.130.538.6+8.1
- AA-Omniscience · Non-hallucinationDeepSeek-R1 0528 by 4.716.611.9+4.7
Long documentso3
- AA-LCRo3 by 1955.774.7+19
- LongBench v2o3 by 6.752.158.8+6.7
Following instructionso3
- IFBencho3 by 31.839.671.4+31.8
- Multi-Challengeo3 by 15.44560.4+15.4
Other results29 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 44.236.580.7+44.2
- HarmBencho3 by 43.854.698.4+43.8
- SimpleQADeepSeek-R1 0528 by 42.992.349.4+42.9
- ARC-AGI-1o3 by 39.621.260.8+39.6
- AA Agentic Indexo3 by 15.320.836.1+15.3
- Artificial Analysis Coding Indexo3 by 14.42438.4+14.4
- AA-Omniscienceo3 by 11.9-27.4-15.6+11.9
- LiveCodeBencho3 by 11.473.384.7+11.4
- TAU-bench (retail)o3 by 1063.973.9+10
- imo_2025o3 by 9.86.816.7+9.8
- Aider-Polygloto3 by 9.771.681.3+9.7
- AA Intelligenceo3 by 7.113.120.2+7.1
- OpenAI-MRCR (128k)o3 by 551.556.5+5
- LiveCodeBench (24/8~25/5)o3 by 2.773.175.8+2.7
- livecodebench_hardo3 by 2.463.666+2.4
- TAU-bench (airline)DeepSeek-R1 0528 by 1.553.552+1.5
- XSTestDeepSeek-R1 0528 by 1.598.897.3+1.5
- bbqo3 by 1.496.597.9+1.4
- AIME 2025DeepSeek-R1 0528 by 1.187.586.4+1.1
- livecodebench_mediumDeepSeek-R1 0528 by 1.190.989.8+1.1
- anthropic_red_teamtie99.198.3tie
- HMMT Feb. 2025tie76.777.5tie
- simple_safety_teststie98.399tie
- ZebraLogictie95.195.8tie
- AIME 2024tie91.491.6tie
- FullStackBenchtie69.469.3tie
- livecodebench_easytie99.299.1tie
- MATH-500 (EM)tie9898.1tie
- MMLU-Protie8585tie
Questions people ask
Which is better, DeepSeek-R1 0528 or o3?
o3 wins all six areas where both have results: coding, agents, reasoning, facts, long documents and following instructions. DeepSeek-R1 0528 wins none, but costs 57% less.
Which is better for coding?
o3. It wins 2 of the 3 coding tests both models report; DeepSeek-R1 0528 wins none, and 1 is a tie.
Which is cheaper?
DeepSeek-R1 0528 costs $1.35 per million input tokens and $3.00 per million output tokens; o3 costs $2.00 and $8.00. That makes DeepSeek-R1 0528 about 57% cheaper for the same work.
How do you compare the two?
We use the 45 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 29 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.