GLM 5.2 vs Kimi K3.1 Pro
Full profiles: GLM 5.2 (Zhipu AI) and Kimi K3.1 Pro (Moonshot AI): every benchmark, price and rank.
Updated · Data from 7 sources
As of 24 Oct 2026, GLM 5.2 (max) and Kimi K3.1 Pro are neck and neck on quality (62.3 vs 66.8 on the BenchLeader Index, inside the ±7.2 margin of error), and GLM 5.2 (max) costs 1.9× less, which makes it the better buy for most workloads. Kimi K3.1 Pro leads in every category we measure both in.
GLM 5.2 benchmarks 
Kimi K3.1 Pro benchmarks Which to pick for…
The pick's numbers come first. Category scores on the BenchLeader scale (50 is the average model, 15 points one standard deviation). High confidence means a lead of half a standard deviation or more; under one point is called either.
| Use case | Pick | Why | Confidence |
|---|---|---|---|
| Agents & tool use | Kimi K3.1 Pro | 67 vs 63 category score; AutomationBench 73.3% vs 28.4% | Medium |
| Reasoning | Kimi K3.1 Pro | 81 vs 70 category score; CritPt 24.6% vs 20.9% | High |
| Knowledge | Kimi K3.1 Pro | 75 vs 54 category score; AA-Omniscience 29.5 vs 4.4 | High |
| Long documents | Kimi K3.1 Pro | 68 vs 64 category score; AA-LCR 86.3% vs 78.3% | Medium |
| Chat quality (human preference) | – | Only GLM 5.2 has an Arena rating (67 category score) | Not comparable |
| Lowest cost | GLM 5.2 | $2.15 vs $4.00 per 1M blended; agentic coding session $0.102 vs $0.160 | – |
| Fastest first answer | Either | 25.2 s (GLM 5.2 (max)) vs 27.7 s (Kimi K3.1 Pro) to the first answer token | – |
| High volume on a budget | Neither flagship | A smaller model is usually the better fit: Cheapest good LLMs, Fastest LLMs | – |
Key differences
- GLM 5.2 (max) and Kimi K3.1 Pro are neck and neck on quality (62.3 vs 66.8, within the ±7.2 margin).
- GLM 5.2 (max) is stronger in coding, human preference, instruction following, maths.
- Kimi K3.1 Pro is stronger in agents & tools, composite, knowledge, long context, reasoning, multimodal.
- GLM 5.2 (max) is 1.9× cheaper ($2.15 vs $4.00 per 1M blended).
At a glance
| Spec | GLM 5.2 | Kimi K3.1 Pro |
|---|---|---|
| Developer | Zhipu AI | Moonshot AI |
| Released | 16 Jun 2026 | 23 Oct 2026 |
| Context window (tokens) | 1,000,000 | 1,048,576 |
| Compared at | GLM 5.2 (max) | Kimi K3.1 Pro |
| Open weights | Yes | No |
| Price in / out / cached input per 1M | $1.40 / $4.40 / $0.260 | $2.00 / $10.00 / $0.200 |
Category by category
Category scores on the BenchLeader scale: 50 is the average evaluated model, 15 points is one standard deviation. A gap under one point is called level.
| Category | GLM 5.2 (max) | Kimi K3.1 Pro | Leader |
|---|---|---|---|
| Reasoning | 69.7 | 81.3 | Kimi K3.1 Pro |
| Coding | 62.3 | – | not comparable |
| Agents & tools | 62.6 | 67.3 | Kimi K3.1 Pro |
| Maths | 57.3 | – | not comparable |
| Knowledge | 53.7 | 74.8 | Kimi K3.1 Pro |
| Instruction following | 71.7 | – | not comparable |
| Human preference | 66.6 | – | not comparable |
| Multimodal | – | 67.0 | not comparable |
| Long context | 64.2 | 68.3 | Kimi K3.1 Pro |
| Composite | 67.2 | 75.8 | Kimi K3.1 Pro |
Where sources disagree
Across the 10 benchmarks both have, GLM 5.2 leads 1, Kimi K3.1 Pro leads 9. Every overall index that rates both puts Kimi K3.1 Pro ahead.
| Overall index | GLM 5.2 (max) | Kimi K3.1 Pro | Leader |
|---|---|---|---|
| BenchLeader Index | 62.3 | 66.8 | neck and neck |
| AA Intelligence Index v4.3.2 | 33.7 | 55.1 | Kimi K3.1 Pro |
Each publisher runs its own tasks, harnesses and effort settings, so their overall verdicts can differ; BenchLeader reconciles them on one scale. † measured at another effort setting of the same model (the publisher did not test the one compared here).
Price, speed and latency
List prices blended at three input tokens per output token. Per-task prices use the token counts on the methodology page; suite cost is what Artificial Analysis paid to run its whole Intelligence Index.
| Measure | GLM 5.2 (max) | Kimi K3.1 Pro |
|---|---|---|
| Blended price per 1M tokens | $2.15 | $4.00 |
| Input / output per 1M | $1.40 / $4.40 | $2.00 / $10.00 |
| Cached input per 1M | $0.260 | $0.200 |
| Price per task: chat reply | $0.0019 | $0.0038 |
| Price per task: summarize a 30-page report | $0.019 | $0.030 |
| Price per task: code edit | $0.015 | $0.027 |
| Price per task: agentic coding session | $0.102 | $0.160 |
| Price per task: structured extraction | $0.0037 | $0.0060 |
| Cost to run a full benchmark suite | $1.47 | $2.11 |
| Output speed | 91 tok/s | 81 tok/s |
| Time to first answer | 25.2 s | 27.7 s |
| Time to a complete answer | 30.7 s | 33.9 s |
Per token is not per task: on Artificial Analysis' full suite GLM 5.2 (max) cost $1.47 and Kimi K3.1 Pro $2.11, because GLM 5.2 (max) used about 1.3× as many tokens for the same work.
Each model is compared at its best-ranked reasoning-effort setting. GLM 5.2 (max) scores 62.3 and takes 25 s to its first answer token; GLM 5.2 scores 52.0 in 0 s, GLM 5.2 (no reasoning) scores 48.0 in 1 s. Kimi K3.1 Pro scores 66.8 and takes 28 s to its first answer token. Higher effort means more thinking before the answer: better scores, longer waits and more tokens billed. How reasoning effort changes scores and cost.
Every benchmark
113 benchmarks, grouped by category; each name links to the board and its leaders. † measured at another effort setting of the same model, shown for reference and not counted as a win.
| Metric | Publisher | GLM 5.2 (max) | Kimi K3.1 Pro |
|---|---|---|---|
| BenchLeader Index | BenchLeader | 62.3 | 66.8 |
| Agents & tools score | BenchLeader | 62.6 | 67.3 |
| Coding score | BenchLeader | 62.3 | – |
| Composite score | BenchLeader | 67.2 | 75.8 |
| Human preference score | BenchLeader | 66.6 | – |
| Instruction following score | BenchLeader | 71.7 | – |
| Knowledge score | BenchLeader | 53.7 | 74.8 |
| Long context score | BenchLeader | 64.2 | 68.3 |
| Maths score | BenchLeader | 57.3 | – |
| Reasoning score | BenchLeader | 69.7 | 81.3 |
| Multimodal score | BenchLeader | – | 67.0 |
| Blended price $/M | BenchLeader | $2.15 | $4.00 |
| Output speed | BenchLeader | 91 tok/s | 81 tok/s |
| Time to first answer | BenchLeader | 25.2 s | 27.7 s |
| Context window | BenchLeader | 1,000,000 tokens | 1,048,576 tokens |
| Reasoning | |||
| GPQA Diamond | Epoch AI Benchmarking Hub | 91.9% | – |
| SimpleBench | SimpleBench | 58.8%† | – |
| Arena Hard Prompts | Arena (LMArena) | 1496 | – |
| LiveBench 2.0 Reasoning | LiveBench | – | 91.1% |
| LiveBench 1.0 Reasoning | LiveBench | 78.6%† | – |
| GPQA Diamond (AA) | Artificial Analysis | 89.5% | – |
| Humanity's Last Exam (AA) | Artificial Analysis | 41.1% | 52.5% |
| GPQA Diamond (Vals) | Vals AI | 85.6%† | – |
| Kagi LLM Benchmark | Kagi LLM Benchmark | 62.6%† | – |
| ARC-AGI-1 | ARC Prize | 77.0%† | – |
| ARC-AGI-2 | ARC Prize | 22.8%† | – |
| CritPt | Artificial Analysis | 20.9% | 24.6% |
| Chess Puzzles | Epoch AI Benchmarking Hub | 22.0% | – |
| Mystery Game Puzzles | Epoch AI Benchmarking Hub | 18.0%† | – |
| LMCA | Epoch AI Benchmarking Hub | 45.8% | – |
| DTBench | Epoch AI Benchmarking Hub | 93.6% | – |
| Coding | |||
| SWE-bench Verified (Epoch) | Epoch AI Benchmarking Hub | 78.7% | – |
| SciCode | SciCode | 51.2% | – |
| WeirdML | WeirdML | 70.1% | – |
| FrontierCode | Cognition | 24.5%† | – |
| Arena Coding | Arena (LMArena) | 1514 | – |
| Arena WebDev | Arena (LMArena) | 1604 | – |
| LiveBench 2.0 Coding | LiveBench | – | 81.8% |
| LiveBench 1.0 Coding | LiveBench | 79.7%† | – |
| SciCode (AA) | Artificial Analysis | 51.2% | 60.8% |
Show all 113 benchmarks
| Metric | Publisher | GLM 5.2 (max) | Kimi K3.1 Pro |
|---|---|---|---|
| Coding | |||
| LiveCodeBench | Vals AI | 69.5%† | – |
| SWE-bench (Vals) | Vals AI | 82.8% | – |
| Code Migration | Vals AI | 37.9% | – |
| ProgramBench | Vals AI | 0.5% | – |
| Vibe Code Bench v1.1 | Vals AI | 64.0% | – |
| SWE Atlas: Codebase QnA | Scale AI SEAL | 48.1%† | – |
| SWE Atlas: Refactoring | Scale AI SEAL | 42.4%† | – |
| SWE Atlas: Test Writing | Scale AI SEAL | 41.5%† | – |
| Surface Evolver Bench | Epoch AI Benchmarking Hub | 55.6%† | – |
| DeepSWE v1.1 | Epoch AI Benchmarking Hub | 43.8% | – |
| ALE-Bench | Epoch AI Benchmarking Hub | 1010.2 | – |
| Terminal-Bench 4.0 (AA) | Artificial Analysis | 1.0% | 67.2% |
| Terminal-Bench 2.1 (AA) | Artificial Analysis | 77.9% | – |
| Startrise Frontend | Startrise | 72.1%† | – |
| FrontierCode 1.1 (AA) | Artificial Analysis | – | 47.1% |
| Agents & tools | |||
| APEX-Agents | Mercor | 45.2%† | – |
| Arena Agent | Arena (LMArena) | 2.1 | – |
| LiveBench Long Agentic | LiveBench | – | 49.5% |
| LiveBench Browser | LiveBench | – | 80.0% |
| LiveBench Agentic Coding | LiveBench | 51.8%† | – |
| Terminal-Bench Hard | Artificial Analysis | 50.8% | – |
| τ²-Bench Telecom (AA) | Artificial Analysis | 99.1% | – |
| Terminal-Bench 2.1 (Vals) | Vals AI | 67.8% | – |
| MCP Atlas | Scale AI SEAL | 77.8%† | – |
| HiL-Bench | Scale AI SEAL | 43.7%† | – |
| τ³-bench Banking | τ²-bench | 37.1%† | – |
| GDPval-AA v2.1 | Artificial Analysis | – | 67.3% |
| τ³-Banking (AA) | Artificial Analysis | 34.6% | – |
| ITBench SRE (AA) | Artificial Analysis | 42.7% | – |
| APEX-Agents (AA) | Artificial Analysis | 33.7% | – |
| Finance Agent v2 | Vals AI | 49.7%† | – |
| Harvey's Legal Agent Benchmark v1 | Vals AI | 7.1% | – |
| Legal Research Bench | Vals AI | 31.3% | – |
| SkillsBench | Vals AI | 45.1% | – |
| SREBench | Vals AI | 0.0% | – |
| PostTrainBench | Epoch AI Benchmarking Hub | 31.7% | – |
| GBAEval | Epoch AI Benchmarking Hub | 0.0%† | – |
| Vending-Bench 2 | Andon Labs | 8313.8† | – |
| AutomationBench | Artificial Analysis | 28.4% | 73.3% |
| GDP.pdf (AA) | Artificial Analysis | 10.4% | 31.4% |
| EnterpriseOps-Gym | Artificial Analysis | 42.7% | – |
| AA-Briefcase v1.1 | Artificial Analysis | – | 1678 |
| CWE-Bench (AA) | Artificial Analysis | – | 60.8% |
| DeepSec-Bench (AA) | Artificial Analysis | – | 36.7% |
| CyberGym end-to-end (AA) | Artificial Analysis | – | 94.7% |
| Maths | |||
| FrontierMath Tiers 1–3 | Epoch AI Benchmarking Hub | 59.2% | – |
| FrontierMath Tier 4 | Epoch AI Benchmarking Hub | 29.3% | – |
| OTIS Mock AIME | Epoch AI Benchmarking Hub | 86.4% | – |
| ProofBench | Vals AI | 35.0% | – |
| LiveBench Mathematics | LiveBench | 89.8%† | – |
| AIME 2026 | MathArena | 90.0%† | – |
| HMMT February 2026 | MathArena | 92.4%† | – |
| MathArena Apex | MathArena | 19.8%† | – |
| Arena Maths | Arena (LMArena) | 1488 | – |
| Knowledge | |||
| SimpleQA Verified | Epoch AI Benchmarking Hub | 34.2% | – |
| LiveBench 2.0 Data Analysis | LiveBench | – | 79.1% |
| LiveBench Research | LiveBench | – | 77.9% |
| LiveBench 1.0 Data Analysis | LiveBench | 73.7%† | – |
| AA-Omniscience | Artificial Analysis | 4.4 | 29.5 |
| MMLU-Pro | Vals AI | 86.7%† | – |
| LegalBench | Vals AI | 84.1%† | – |
| CorpFin | Vals AI | 66.1%† | – |
| TaxEval | Vals AI | 73.3%† | – |
| Excel Modeling Benchmark | Vals AI | 61.5%† | – |
| MedCode | Vals AI | 40.8%† | – |
| MedScribe | Vals AI | 83.5%† | – |
| DrugDiscoveryBench | Scale AI SEAL | 36.2%† | – |
| EBR-bench | Epoch AI Benchmarking Hub | 9.5% | – |
| AA-Omniscience: accuracy | Artificial Analysis | 24.3% | 51.0% |
| AA-Omniscience: non-hallucination | Artificial Analysis | 73.7% | 56.1% |
| Instruction following | |||
| LiveBench Language | LiveBench | 76.2%† | – |
| LiveBench Instruction Following | LiveBench | 62.3%† | – |
| IFBench | Artificial Analysis | 73.3% | – |
| Arena Instruction Following | Arena (LMArena) | 1469 | – |
| Human preference | |||
| Arena Text | Arena (LMArena) | 1475 | – |
| EQ-Bench 4 | EQ-Bench | 1222† | – |
| Arena Creative Writing | Arena (LMArena) | 1453 | – |
| Arena Multi-turn | Arena (LMArena) | 1475 | – |
| Arena Longer Queries | Arena (LMArena) | 1486 | – |
| Multimodal | |||
| LiveBench Design | LiveBench | – | 77.8% |
| MMMU-Pro | Artificial Analysis | – | 83.9% |
| Long context | |||
| AA-LCR | Artificial Analysis | 78.3% | 86.3% |
| MLCR | Artificial Analysis | 12.8% | – |
| Composite | |||
| Epoch Capabilities Index | Epoch AI Benchmarking Hub | 151.9† | – |
| LiveBench 2.0 | LiveBench | – | 76.7% |
| LiveBench 1.0 | LiveBench | 73.2%† | – |
| AA Intelligence Index v4.3.2 | Artificial Analysis | 33.7 | 55.1 |
| AA Cyber Index | Artificial Analysis | – | 64.1 |
Data as of 2026-10-24. Best configuration of each model; every score links to its source on the model pages.
What changed
- Kimi K3.1 Pro: BenchLeader Index 65.6 → 66.8.
- GLM 5.2 (max): cached input price moved from $0.30/M to $0.26/M (13% down).
- Kimi K3.1 Pro: 5 new results (FrontierCode 1.1 (AA), AA Cyber Index, CWE-Bench (AA)…).
- GLM 5.2 (max): BenchLeader Index 62.8 → 62.3.
- GLM 5.2 (max): 1 new result (Harvey's Legal Agent Benchmark v1).
- Kimi K3.1 Pro: first results.
- GLM 5.2 (max): 1 score revision.
- GLM 5.2 (max): cached input price moved from $0.26/M to $0.30/M (15% up).
- GLM 5.2 (max): 4 new results (AutomationBench, GDP.pdf (AA), MLCR…).
- GLM 5.2 (max): 1 new result (Harvey's Legal Agent Benchmark v1.1).
Related comparisons
- GLM 5.2 vs Claude Opus 5.5
- GLM 5.2 vs Gemini 4 Argon
- GLM 5.2 vs Muse Wonder Full
- Kimi K3.1 Pro vs Claude Opus 5.5
- Kimi K3.1 Pro vs Gemini 4 Argon
- Kimi K3.1 Pro vs Muse Wonder Full
- Kimi K3.1 Pro vs Claude Fable 5.1
- GLM 5.2 vs Kimi K3
Rankings: Best LLM overall · Best LLM for agents · Best cheap LLMs
GLM 5.2 vs Kimi K3.1 Pro: questions
- Is GLM 5.2 better than Kimi K3.1 Pro?
- GLM 5.2 (max) and Kimi K3.1 Pro are neck and neck on quality (62.3 vs 66.8, within the ±7.2 margin). The BenchLeader Index combines every independent quality benchmark; the two are within the margin of error as of 2026-10-24, so check the category scores for your use.
- Which is cheaper, GLM 5.2 or Kimi K3.1 Pro, and will that change?
- GLM 5.2 is cheaper today: $2.15 against $4.00 per million tokens, blended at three input tokens per output token. No price change has been announced for either.
- Is GLM 5.2 better than Kimi K3.1 Pro for agents and tool use?
- Kimi K3.1 Pro scores higher in agentic tasks (67 vs 63 on the category index, where 50 is average).
- Which is better for long documents?
- Kimi K3.1 Pro scores higher on long-context benchmarks (68 vs 64). Both take about 1M tokens: GLM 5.2 1,000,000, Kimi K3.1 Pro 1,048,576.
- Which has lower latency?
- GLM 5.2 (max) answers first: 25.2 s against 27.7 s to the first answer token, as Artificial Analysis measures it. Reasoning effort drives most of this: a higher setting thinks longer before it answers.
- Which is faster, GLM 5.2 or Kimi K3.1 Pro?
- GLM 5.2 streams faster: 91 against 81 output tokens per second.
- Do independent benchmarks agree on GLM 5.2 vs Kimi K3.1 Pro?
- Across the 10 benchmarks both have, GLM 5.2 leads 1, Kimi K3.1 Pro leads 9. Every overall index that rates both puts Kimi K3.1 Pro ahead.
- When was Kimi K3.1 Pro released?
- Kimi K3.1 Pro was released on 23 Oct 2026.