DeepSeek-R1 vs DeepSeek-R1 0528
Wins 1 of 6 areas
Long documents
Wins 5 of 6 areas
Coding · Agents · Reasoning · Facts · Following instructions
DeepSeek-R1 0528 is the stronger all-rounder.DeepSeek-R1 is better at long documents.
Scores updated · 39 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 thinking02DeepSeek-R1 05282 of 3 tests · 1 tie
- CodingWriting and fixing software12DeepSeek-R1 05282 of 3 tests
- FactsGetting facts right instead of making them up01DeepSeek-R1 05281 of 2 tests · 1 tie
- Following instructionsDoing exactly what it is asked01DeepSeek-R1 05281 of 2 tests · 1 tie
- AgentsCarrying out multi-step tasks on its own01DeepSeek-R1 05281 of 1 test
- Long documentsFinding answers in very long texts20DeepSeek-R12 of 2 tests
Agents rests on a single test.
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 28% less for the same work.
The biggest differences
The three tests each model wins by the widest margin. Scores are out of 100.
Where DeepSeek-R1 0528 pulls ahead
- Graduate-level biology, physics and chemistry questionsGPQA Diamond+10.5points ahead
- Hard command-line tasks in a real terminalTerminal-Bench Hard+9.8points ahead
- Very hard expert questions across many subjectsHumanity's Last Exam+7.3points ahead
Where DeepSeek-R1 pulls ahead
- Questions about very long textsLongBench v2+6.2points ahead
- Fixes real GitHub issues in Python projectsSWE-bench Verified+4.6points ahead
- Reasons across sets of long documentsAA-LCR+2points ahead
Every test, side by side
All 39 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingDeepSeek-R1 0528
- Terminal-Bench HardDeepSeek-R1 0528 by 9.86.115.9+9.8
- SWE-bench VerifiedDeepSeek-R1 by 4.649.244.6+4.6
- SciCodeDeepSeek-R1 0528 by 238.340.3+2
ReasoningDeepSeek-R1 0528
- GPQA DiamondDeepSeek-R1 0528 by 10.570.881.3+10.5
- Humanity's Last ExamDeepSeek-R1 0528 by 7.38.515.8+7.3
- CritPttie0.61.4tie
FactsDeepSeek-R1 0528
- AA-Omniscience · Non-hallucinationDeepSeek-R1 0528 by 6.99.716.6+6.9
- AA-Omniscience · Accuracytie30.530.5tie
Long documentsDeepSeek-R1
- LongBench v2DeepSeek-R1 by 6.258.352.1+6.2
- AA-LCRDeepSeek-R1 by 257.755.7+2
Following instructionsDeepSeek-R1 0528
- Multi-ChallengeDeepSeek-R1 0528 by 4.340.745+4.3
- IFBenchtie3939.6tie
Other results26 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.
- SimpleQADeepSeek-R1 0528 by 62.230.192.3+62.2
- Codeforces-Div1 (Rating)DeepSeek-R1 0528 by 400 rating points15301930+400 rating
- HMMT Feb. 2025DeepSeek-R1 0528 by 3541.776.7+35
- usamo_2025DeepSeek-R1 0528 by 25.34.830.1+25.3
- τ²-Bench Telecom (AA run)DeepSeek-R1 0528 by 25.111.436.5+25.1
- Aider-PolyglotDeepSeek-R1 0528 by 18.353.371.6+18.3
- AIME 2025DeepSeek-R1 0528 by 17.57087.5+17.5
- LiveCodeBench (24/8~25/5)DeepSeek-R1 0528 by 17.255.973.1+17.2
- ZebraLogicDeepSeek-R1 0528 by 16.478.795.1+16.4
- OpenAI-MRCR (128k)DeepSeek-R1 0528 by 15.735.851.5+15.7
- AIME 2024DeepSeek-R1 0528 by 11.679.891.4+11.6
- LiveCodeBenchDeepSeek-R1 0528 by 9.863.573.3+9.8
- HarmBenchDeepSeek-R1 0528 by 6.747.954.6+6.7
- ARC-AGI-1DeepSeek-R1 0528 by 5.415.821.2+5.4
- AA-OmniscienceDeepSeek-R1 0528 by 4.8-32.2-27.4+4.8
- XSTestDeepSeek-R1 0528 by 4.494.498.8+4.4
- anthropic_red_teamDeepSeek-R1 0528 by 1.997.299.1+1.9
- AA IntelligenceDeepSeek-R1 0528 by 1.711.413.1+1.7
- MMLU-ProDeepSeek-R1 0528 by 18485+1
- FullStackBenchtie70.169.4tie
- MATH-500 (EM)tie97.398tie
- Artificial Analysis Coding Indextie24.624tie
- FRAMES (Acc.)tie82.583tie
- MMLU-Reduxtie92.993.4tie
- simple_safety_teststie9898.3tie
- bbqtie96.696.5tie
Questions people ask
Which is better, DeepSeek-R1 or DeepSeek-R1 0528?
DeepSeek-R1 0528 wins five of the six areas where both have results: coding, agents, reasoning, facts and following instructions. DeepSeek-R1 wins long documents.
Which is better for coding?
DeepSeek-R1 0528. It wins 2 of the 3 coding tests both models report; DeepSeek-R1 wins 1.
Which is cheaper?
DeepSeek-R1 costs $2.00 per million input tokens and $4.00 per million output tokens; DeepSeek-R1 0528 costs $1.35 and $3.00. That makes DeepSeek-R1 0528 about 28% cheaper for the same work.
How do you compare the two?
We use the 39 benchmark tests both models have published scores on. The verdict counts the 13 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 26 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.