DeepSeek-R1 vs GLM-4.5-Air
Wins 3 of 6 areas
Facts · Long documents · Following instructions
Wins 2 of 6 areas
Coding · Agents
DeepSeek-R1 wins more areas, narrowly.GLM-4.5-Air is cheaper and better at coding.
Scores updated · 28 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.
- FactsGetting facts right instead of making them up21DeepSeek-R12 of 3 tests
- Long documentsFinding answers in very long texts10DeepSeek-R11 of 1 test
- Following instructionsDoing exactly what it is asked10DeepSeek-R11 of 1 test
- CodingWriting and fixing software12GLM-4.5-Air2 of 3 tests
- AgentsCarrying out multi-step tasks on its own01GLM-4.5-Air1 of 1 test
- ReasoningHard problems that need careful thinking11Even1 each · 1 tie
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.
GLM-4.5-Air costs 81% 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 pulls ahead
- Answers hard knowledge questions correctlyAA-Omniscience · Accuracy+14.2points ahead
- Reasons across sets of long documentsAA-LCR+11points ahead
- Code for real scientific research problemsSciCode+7.7points ahead
Where GLM-4.5-Air pulls ahead
- Hard command-line tasks in a real terminalTerminal-Bench Hard+14.4points ahead
- Fixes real GitHub issues in Python projectsSWE-bench Verified+8.4points ahead
- Graduate-level biology, physics and chemistry questionsGPQA Diamond+2.5points ahead
Every test, side by side
All 28 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingGLM-4.5-Air
- Terminal-Bench HardGLM-4.5-Air by 14.46.120.5+14.4
- SWE-bench VerifiedGLM-4.5-Air by 8.449.257.6+8.4
- SciCodeDeepSeek-R1 by 7.738.330.6+7.7
ReasoningEven
- GPQA DiamondGLM-4.5-Air by 2.570.873.3+2.5
- Humanity's Last ExamDeepSeek-R1 by 1.58.57+1.5
- CritPttie0.60tie
FactsDeepSeek-R1
- AA-Omniscience · AccuracyDeepSeek-R1 by 14.230.516.3+14.2
- AA-Omniscience · Non-hallucinationDeepSeek-R1 by 2.69.77.1+2.6
- Vectara HHEM hallucination ratelower is betterGLM-4.5-Air by 211.39.3+2
Following instructionsDeepSeek-R1
- IFBenchDeepSeek-R1 by 1.43937.6+1.4
Other results16 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)GLM-4.5-Air by 35.111.446.5+35.1
- AA-OmniscienceDeepSeek-R1 by 29.3-32.2-61.5+29.3
- vectara_avg_summary_lengthDeepSeek-R1 by 22.993.570.6+22.9
- AIME 2025GLM-4.5-Air by 13.37083.3+13.3
- HarmBenchGLM-4.5-Air by 8.347.956.1+8.3
- LiveCodeBenchGLM-4.5-Air by 7.263.570.7+7.2
- AIR-Bench 2024GLM-4.5-Air by 4.252.957.1+4.2
- XSTestGLM-4.5-Air by 4.294.498.6+4.2
- MMLU-ProDeepSeek-R1 by 2.68481.4+2.6
- vectara_factual_consistencyGLM-4.5-Air by 288.790.7+2
- anthropic_red_teamGLM-4.5-Air by 1.897.299+1.8
- vectara_answer_rateGLM-4.5-Air by 1.19798.1+1.1
- simple_safety_testsGLM-4.5-Air by 19899+1
- Artificial Analysis Coding Indextie24.623.8tie
- MATH-500 (EM)tie97.398.1tie
- AA Intelligencetie11.411.1tie
Questions people ask
Which is better, DeepSeek-R1 or GLM-4.5-Air?
DeepSeek-R1 wins three of the six areas where both have results: facts, long documents and following instructions. GLM-4.5-Air wins coding and agents, and costs 81% less. They are level on reasoning.
Which is better for coding?
GLM-4.5-Air. 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; GLM-4.5-Air costs $0.17 and $0.98. That makes GLM-4.5-Air about 81% cheaper for the same work.
How do you compare the two?
We use the 28 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 16 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.