DeepSeek-R1 0528 vs GLM-4.5
Wins 2 of 6 areas
Reasoning · Long documents
Wins 2 of 6 areas
Coding · Following instructions
The two are evenly matched.DeepSeek-R1 0528 is better at reasoning and long documents; GLM-4.5 at coding and following instructions.
Scores updated · 24 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 thinking30DeepSeek-R1 05283 of 3 tests
- Long documentsFinding answers in very long texts10DeepSeek-R1 05281 of 1 test
- CodingWriting and fixing software12GLM-4.52 of 3 tests
- Following instructionsDoing exactly what it is asked01GLM-4.51 of 1 test
- AgentsCarrying out multi-step tasks on its own11Even1 each
- FactsGetting facts right instead of making them up11Even1 each
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.
GLM-4.5 costs 36% 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
- Code for real scientific research problemsSciCode+5.5points ahead
- Answers hard knowledge questions correctlyAA-Omniscience · Accuracy+5.4points ahead
- Graduate-level biology, physics and chemistry questionsGPQA Diamond+3.1points ahead
Where GLM-4.5 pulls ahead
- Fixes real GitHub issues in Python projectsSWE-bench Verified+19.6points ahead
- Finds hard-to-locate facts by browsing the webBrowseComp+17.5points ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+13.3points ahead
Every test, side by side
All 24 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingGLM-4.5
- SWE-bench VerifiedGLM-4.5 by 19.644.664.2+19.6
- Terminal-Bench HardGLM-4.5 by 6.115.922+6.1
- SciCodeDeepSeek-R1 0528 by 5.540.334.8+5.5
ReasoningDeepSeek-R1 0528
- GPQA DiamondDeepSeek-R1 0528 by 3.181.378.2+3.1
- Humanity's Last ExamDeepSeek-R1 0528 by 2.815.813+2.8
- CritPtDeepSeek-R1 0528 by 1.41.40+1.4
FactsEven
- AA-Omniscience · Non-hallucinationGLM-4.5 by 13.316.629.9+13.3
- AA-Omniscience · AccuracyDeepSeek-R1 0528 by 5.430.525.1+5.4
Other results12 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.
- Terminal-BenchGLM-4.5 by 31.85.737.5+31.8
- TAU-bench (retail)GLM-4.5 by 15.863.979.7+15.8
- TAU-bench (airline)GLM-4.5 by 6.953.560.4+6.9
- τ²-Bench Telecom (AA run)GLM-4.5 by 6.536.543+6.5
- AIME 2025GLM-4.5 by 5.887.593.3+5.8
- AA Agentic IndexDeepSeek-R1 0528 by 4.620.816.2+4.6
- Artificial Analysis Coding IndexGLM-4.5 by 2.32426.3+2.3
- LiveCodeBenchtie73.372.9tie
- MMLU-Protie8584.6tie
- AA Intelligencetie13.112.8tie
- MATH-500 (EM)tie9898.2tie
- AA-Omnisciencetie-27.4-27.4tie
Questions people ask
Which is better, DeepSeek-R1 0528 or GLM-4.5?
DeepSeek-R1 0528 and GLM-4.5 each win two of the six areas where both have results. DeepSeek-R1 0528 wins reasoning and long documents; GLM-4.5 wins coding and following instructions. GLM-4.5 costs 36% less. They are level on agents and facts.
Which is better for coding?
GLM-4.5. It wins 2 of the 3 coding tests both models report; DeepSeek-R1 0528 wins 1.
Which is cheaper?
DeepSeek-R1 0528 costs $1.35 per million input tokens and $3.00 per million output tokens; GLM-4.5 costs $0.60 and $2.20. That makes GLM-4.5 about 36% cheaper for the same work.
How do you compare the two?
We use the 24 benchmark tests both models have published scores on. The verdict counts the 12 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 12 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.