GLM-5 vs Kimi K3
Full profiles: GLM-5 (Zhipu AI) and Kimi K3 (Moonshot AI): every benchmark, price and rank.
Updated · Data from 12 sources
As of 2 Nov 2026, GLM-5 and Kimi K3 are neck and neck on quality (60.2 vs 62.1 on the BenchLeader Index, inside the ±4.7 margin), and GLM-5 is 3.9× cheaper at $1.55 vs $6.00 per 1M tokens blended. Choose GLM-5 for reasoning (68 vs 60); choose Kimi K3 for long context (69 vs 63), coding (64 vs 59), and knowledge (60 vs 57).
Key facts (as of 2 Nov 2026)
- GLM-5 scores 60.2 on the BenchLeader Index and Kimi K3 scores 62.1 (BenchLeader, 2 Nov 2026).
- GLM-5 costs $1.55 per 1M tokens blended (list price, 2 Nov 2026).
- Kimi K3 costs $6.00 per 1M tokens blended (list price, 2 Nov 2026).
- Of the 11 benchmarks both models have, Kimi K3 leads 11 (BenchLeader, 2 Nov 2026).
GLM-5 benchmarks 
Kimi K3 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 |
|---|---|---|---|
| Coding | Kimi K3 | 64 vs 59 category score; LiveCodeBench 87.2% vs 81.9% | Medium |
| Agents & tool use | Kimi K3 | 59 vs 58 category score; τ³-bench Banking 37.1% vs 9.8% | Medium |
| Maths | Either | 63 vs 63 category score | Either |
| Reasoning | GLM-5 | 68 vs 60 category score; GPQA Diamond (AA) 82.0% vs 93.5% | High |
| Knowledge | Kimi K3 | 60 vs 57 category score; MMLU-Pro 88.0% vs 86.0% | Medium |
| Long documents | Kimi K3 | 69 vs 63 category score; AA-LCR v1.1 88.7% vs 75.7% | Medium |
| Chat quality (human preference) | – | Only Kimi K3 has an Arena rating (68 category score) | Not comparable |
| Lowest cost | GLM-5 | $1.55 vs $6.00 per 1M blended; agentic coding session $0.073 vs $0.240 | – |
| Fastest first answer | Kimi K3 | 37.2 s (Kimi K3 (max)) vs 51.4 s (GLM-5 (thinking)) 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 (thinking) and Kimi K3 (max) are neck and neck on quality (60.2 vs 62.1, within the ±4.7 margin).
- GLM-5 (thinking) is stronger in instruction following, reasoning.
- Kimi K3 (max) is stronger in agents & tools, coding, knowledge, long context, human preference, multimodal.
- GLM-5 (thinking) is 3.9× cheaper ($1.55 vs $6.00 per 1M blended).
At a glance
| Spec | GLM-5 | Kimi K3 |
|---|---|---|
| Developer | Zhipu AI | Moonshot AI |
| Released | 11 Feb 2026 | 16 Jul 2026 |
| Context window (tokens) | 204,800 | 1,048,576 |
| Compared at | GLM-5 (thinking) | Kimi K3 (max) |
| Open weights | Yes | Yes |
| Price in / out / cached input per 1M | $1.00 / $3.20 / $0.200 | $3.00 / $15.00 / $0.300 |
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 (thinking) | Kimi K3 (max) | Leader |
|---|---|---|---|
| Reasoning | 68.2 | 60.0 | GLM-5 |
| Coding | 59.1 | 64.4 | Kimi K3 |
| Agents & tools | 58.0 | 59.2 | Kimi K3 |
| Maths | 63.2 | 63.2 | level |
| Knowledge | 57.3 | 59.8 | Kimi K3 |
| Instruction following | 70.8 | – | not comparable |
| Human preference | – | 68.0 | not comparable |
| Multimodal | – | 63.1 | not comparable |
| Long context | 62.6 | 69.4 | Kimi K3 |
On one chart
Each domain on the index's own scale, where 50 is an average model. Click a domain to see the benchmarks behind it. Pick up to 5 models from the list.
Not drawn, measured for fewer than two of these models: Instruction following, Multimodal, Human preference.
ModelIndex
Where sources disagree
Across the 11 benchmarks both have, GLM-5 leads 0, Kimi K3 leads 11. Every overall index that rates both puts Kimi K3 ahead.
| Overall index | GLM-5 (thinking) | Kimi K3 (max) | Leader |
|---|---|---|---|
| BenchLeader Index | 60.2 | 62.1 | neck and neck |
| Arena Text | 1457† | 1488 | Kimi K3 |
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. Cost, tokens and time per benchmark task are measured: Artificial Analysis’ averages over its whole Intelligence Index run.
| Measure | GLM-5 (thinking) | Kimi K3 (max) |
|---|---|---|
| Blended price per 1M tokens | $1.55 | $6.00 |
| Input / output per 1M | $1.00 / $3.20 | $3.00 / $15.00 |
| Cached input per 1M | $0.200 | $0.300 |
| Price per task: chat reply | $0.0014 | $0.0057 |
| Price per task: summarize a 30-page report | $0.014 | $0.045 |
| Price per task: code edit | $0.011 | $0.041 |
| Price per task: agentic coding session | $0.073 | $0.240 |
| Price per task: structured extraction | $0.0026 | $0.0090 |
| Agentic coding session, input mostly cached | $0.037 | $0.118 |
| Output speed | 62 tok/s | 66 tok/s |
| Time to first answer | 51.4 s | 37.2 s |
| Time to a complete answer | 59.4 s | 44.8 s |
Each model is compared at its best-ranked reasoning-effort setting. GLM-5 (thinking) scores 60.2 and takes 51 s to its first answer token; GLM-5 scores 54.9 in 2 s. Kimi K3 (max) scores 62.1 and takes 37 s to its first answer token; Kimi K3 scores 59.1 in 1 s, Kimi K3 (low) scores 53.8 in 40 s. 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
159 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 (thinking) | Kimi K3 (max) |
|---|---|---|---|
| BenchLeader Index | BenchLeader | 60.2 | 62.1 |
| Agents & tools score | BenchLeader | 58.0 | 59.2 |
| Coding score | BenchLeader | 59.1 | 64.4 |
| Instruction following score | BenchLeader | 70.8 | – |
| Knowledge score | BenchLeader | 57.3 | 59.8 |
| Long context score | BenchLeader | 62.6 | 69.4 |
| Maths score | BenchLeader | 63.2 | 63.2 |
| Reasoning score | BenchLeader | 68.2 | 60.0 |
| Human preference score | BenchLeader | – | 68.0 |
| Multimodal score | BenchLeader | – | 63.1 |
| Blended price $/M | BenchLeader | $1.55 | $6.00 |
| Output speed | BenchLeader | 62 tok/s | 66 tok/s |
| Time to first answer | BenchLeader | 51.4 s | 37.2 s |
| Context window | BenchLeader | 204,800 tokens | 1,048,576 tokens |
| Reasoning | |||
| GPQA Diamond | Epoch AI Benchmarking Hub | 87.8%† | 93.1% |
| SimpleBench | SimpleBench | 53.2%† | 60.7% |
| Arena Hard Prompts | Arena (LMArena) | 1478† | 1517 |
| LiveBench 2.0 Reasoning | LiveBench | – | 90.7%† |
| LiveBench 1.0 Reasoning | LiveBench | – | 90.7%† |
| GPQA Diamond (AA) | Artificial Analysis | 82.0% | 93.5% |
| Humanity's Last Exam (AA) | Artificial Analysis | 29.3% | 46.9% |
| HLE-Diamond (AA) | Artificial Analysis | – | 26.9% |
| GPQA Diamond (Vals) | Vals AI | 83.3% | 92.9%† |
| Kagi LLM Benchmark | Kagi LLM Benchmark | 75.0% | – |
| ARC-AGI-1 | ARC Prize | 44.7%† | 94.5% |
| ARC-AGI-2 | ARC Prize | 4.9%† | 60.4% |
| CritPt | Artificial Analysis | 2.0% | 23.4% |
| CritPt v2 hard subset (AA) | Artificial Analysis | – | 39.5% |
| Chess Puzzles | Epoch AI Benchmarking Hub | – | 36.0% |
| Mystery Game Puzzles | Epoch AI Benchmarking Hub | – | 26.0% |
| CL-bench | Epoch AI Benchmarking Hub | 18.7%† | – |
| LMCA | Epoch AI Benchmarking Hub | – | 52.7% |
| DTBench | Epoch AI Benchmarking Hub | – | 91.2% |
| ForecastBench | ForecastBench | 63.7%† | 64.7% |
| Coding | |||
| SWE-bench Verified (Epoch) | Epoch AI Benchmarking Hub | 72.1%† | – |
| SciCode | SciCode | – | 59.5% |
| WeirdML | WeirdML | 48.2%† | 82.6% |
| FrontierCode | Cognition | – | 44.2%† |
| Arena Coding | Arena (LMArena) | 1498† | 1540 |
Show all 159 benchmarks
| Metric | Publisher | GLM-5 (thinking) | Kimi K3 (max) |
|---|---|---|---|
| Coding | |||
| Arena WebDev | Arena (LMArena) | 1436† | 1656 |
| LiveBench 2.0 Coding | LiveBench | – | 59.3%† |
| LiveBench 1.0 Coding | LiveBench | – | 81.5%† |
| LiveCodeBench | Vals AI | 81.9% | 87.2% |
| IOI | Vals AI | – | 48.9% |
| SWE-bench (Vals) | Vals AI | 71.4% | 93.4%† |
| SWE-bench Verified (bash only) | SWE-bench | 72.8%† | – |
| SWE-bench Verified (any scaffold) | SWE-bench | 72.8%† | – |
| Code Migration | Vals AI | – | 16.1% |
| Code Migration v1.1 | Vals AI | – | 28.6% |
| ProgramBench | Vals AI | – | 2.0%† |
| SWE-bench Multimodal | Vals AI | – | 44.2% |
| Vibe Code Bench 1-100 | Vals AI | – | 18.2% |
| Vibe Code Bench v1.1 | Vals AI | 23.4% | 85.0% |
| SWE Atlas: Codebase QnA | Scale AI SEAL | 20.5%† | – |
| SWE Atlas: Refactoring | Scale AI SEAL | 24.2%† | – |
| SWE Atlas: Test Writing | Scale AI SEAL | 28.7%† | – |
| Surface Evolver Bench | Epoch AI Benchmarking Hub | – | 93.0% |
| DeepSWE v1.1 | Epoch AI Benchmarking Hub | – | 68.5% |
| ALE-Bench | Epoch AI Benchmarking Hub | 765.6† | 1524.5 |
| FrontierSWE | Epoch AI Benchmarking Hub | – | 25.9% |
| Terminal-Bench 4.0 (AA) | Artificial Analysis | – | 12.6% |
| Terminal-Bench 2.1 (AA) | Artificial Analysis | – | 85.0% |
| Startrise Frontend | Startrise | – | 79.8%† |
| GameBench 2 Overall | DuelLab GameBench 2 | – | 48.8† |
| GameBench 2 by reasoning setting | DuelLab GameBench 2 | – | 45.1 |
| FrontierCode 1.1 (AA) | Artificial Analysis | – | 43.6% |
| FrontierCode 2.0 (AA) | Artificial Analysis | – | 10.9%† |
| Agents & tools | |||
| Terminal-Bench | Terminal-Bench | 52.4%† | – |
| APEX-Agents | Mercor | – | 47.8%† |
| Arena Agent | Arena (LMArena) | – | 1.1 |
| LiveBench Long Agentic | LiveBench | – | 26.4%† |
| LiveBench Browser | LiveBench | – | 75.2%† |
| LiveBench Agentic Coding | LiveBench | – | 62.2%† |
| Terminal-Bench Hard | Artificial Analysis | 43.2% | – |
| τ²-Bench Telecom (AA) | Artificial Analysis | 98.3% | – |
| Terminal-Bench 2.1 (Vals) | Vals AI | – | 80.9%† |
| MCP Atlas | Scale AI SEAL | – | 82.3% |
| HiL-Bench | Scale AI SEAL | – | 42.7%† |
| τ²-bench | τ²-bench | 81.0% | – |
| τ³-bench Banking | τ²-bench | 9.8% | 37.1% |
| τ³-Banking (AA) | Artificial Analysis | – | 46.0% |
| ITBench SRE (AA) | Artificial Analysis | – | 47.7% |
| Analyst Agent (AA) | Artificial Analysis | – | 38.8% |
| APEX-Agents (AA) | Artificial Analysis | 14.4% | 41.3% |
| CyberBench | Vals AI | – | 75.2% |
| Finance Agent v2 | Vals AI | – | 53.1% |
| Finance Agent v2.1 | Vals AI | – | 60.0% |
| Harvey's Legal Agent Benchmark v1.1 | Vals AI | – | 12.9% |
| Harvey's Legal Agent Benchmark v1 | Vals AI | – | 12.9% |
| Legal Research Bench | Vals AI | – | 46.1% |
| Tax Agent Bench | Vals AI | – | 33.4% |
| Terminal-Bench 2.0 (Vals) | Vals AI | 49.4% | – |
| Terminal-Bench 4.0 (Vals) | Vals AI | – | 17.2% |
| Terminal-Bench Science | Vals AI | – | 1.4% |
| Time Horizon Index: KSP | Vals AI | – | 10.5% |
| PostTrainBench | Epoch AI Benchmarking Hub | – | 32.0%† |
| GBAEval | Epoch AI Benchmarking Hub | – | 48.4%† |
| Vending-Bench 2 | Andon Labs | 4432.1† | 5165.0† |
| AutomationBench | Artificial Analysis | – | 58.3% |
| GDP.pdf (AA) | Artificial Analysis | – | 22.0% |
| Harvey LAB-AA v1.1 | Harvey | – | 5.3% |
| EnterpriseOps-Gym | Artificial Analysis | – | 45.3% |
| AA-Briefcase v1.1 | Artificial Analysis | – | 1498 |
| GDPx | GDPx by SemiAnalysis | – | 3.7% |
| Terminal-Bench-Science 0.1 (AA) | Artificial Analysis | – | 8.1% |
| CWE-Bench (AA) | Artificial Analysis | – | 53.3% |
| DeepSec-Bench (AA) | Artificial Analysis | – | 12.4% |
| CyberGym end-to-end (AA) | Artificial Analysis | – | 58.0% |
| Maths | |||
| FrontierMath Tiers 1–3 | Epoch AI Benchmarking Hub | – | 72.2% |
| FrontierMath Tier 4 | Epoch AI Benchmarking Hub | – | 39.0% |
| OTIS Mock AIME | Epoch AI Benchmarking Hub | 80.0%† | 97.2% |
| ProofBench | Vals AI | – | 87.0%† |
| LiveBench Mathematics | LiveBench | – | 84.4%† |
| AIME (Vals) | Vals AI | 91.7% | – |
| AIME 2026 | MathArena | 96.7%† | 96.7%† |
| HMMT February 2026 | MathArena | 86.4%† | 97.0%† |
| USAMO 2026 | MathArena | 35.1%† | – |
| MathArena Apex | MathArena | 10.9%† | 65.6%† |
| Arena Maths | Arena (LMArena) | 1444† | 1499 |
| Knowledge | |||
| SimpleQA Verified | Epoch AI Benchmarking Hub | – | 50.6% |
| LiveBench 2.0 Data Analysis | LiveBench | – | 78.7%† |
| LiveBench Research | LiveBench | – | 79.0%† |
| LiveBench 1.0 Data Analysis | LiveBench | – | 78.7%† |
| AA-Omniscience v1.1 | Artificial Analysis | – | -1.7† |
| MMLU-Pro | Vals AI | 86.0% | 88.0% |
| LegalBench | Vals AI | 84.1% | 86.2% |
| CorpFin | Vals AI | 62.9% | 71.6%† |
| TaxEval | Vals AI | 70.0% | 75.7%† |
| MedQA | Vals AI | 94.3% | – |
| BioMysteryBench | Vals AI | – | 72.6% |
| CaseLaw v2 | Vals AI | 52.5% | – |
| Excel Modeling Benchmark | Vals AI | – | 66.7% |
| Excel Modeling Benchmark v1.1 | Vals AI | – | 71.3% |
| MedCode | Vals AI | – | 49.4% |
| MedScribe | Vals AI | – | 88.0% |
| Public Benefits Bench v1.1 | Vals AI | – | 68.2% |
| GDP.pdf | Epoch AI Benchmarking Hub | – | 19.0% |
| AA-Omniscience v1.1: accuracy | Artificial Analysis | – | 42.4%† |
| AA-Omniscience v1.1: non-hallucination | Artificial Analysis | – | 23.4%† |
| HealthBench Professional (AA) | OpenAI | – | 53.2% |
| Instruction following | |||
| LiveBench Language | LiveBench | – | 85.5%† |
| LiveBench Instruction Following | LiveBench | – | 71.4%† |
| IFBench | Artificial Analysis | 72.3% | – |
| Arena Instruction Following | Arena (LMArena) | 1447† | 1490 |
| Human preference | |||
| Arena Text | Arena (LMArena) | 1457† | 1488 |
| EQ-Bench 4 | EQ-Bench | – | 1319† |
| Arena Creative Writing | Arena (LMArena) | 1448† | 1463 |
| Arena Multi-turn | Arena (LMArena) | 1471† | 1502 |
| Arena Longer Queries | Arena (LMArena) | 1468† | 1503 |
| Multimodal | |||
| LiveBench Design | LiveBench | – | 77.2%† |
| MMMU-Pro | Artificial Analysis | – | 80.5% |
| MMMU-Pro (Vals) | Vals AI | – | 88.2%† |
| MortgageTax | Vals AI | – | 66.3%† |
| SAGE | Vals AI | – | 86.8% |
| Blueprint-Bench 2 | Epoch AI Benchmarking Hub | – | 29.5%† |
| Code2Video-AA | Artificial Analysis | – | 1041 |
| AA-3D | Artificial Analysis | – | 982 |
| Long context | |||
| AA-LCR v1.1 | Artificial Analysis | 75.7% | 88.7% |
| MLCR | Artificial Analysis | – | 21.7% |
| Composite | |||
| Epoch Capabilities Index | Epoch AI Benchmarking Hub | 145.8† | 158.3† |
| LiveBench 2.0 | LiveBench | – | 64.6%† |
| LiveBench 1.0 | LiveBench | – | 79.2%† |
| AA Intelligence Index v5.0 | Artificial Analysis | – | 17.4† |
| AA Intelligence Index v4.3.2 | Artificial Analysis | 27.9 | 43.6 |
| AA Progress Index v1 | Artificial Analysis | 280.9 | 319.9 |
| Vals Index | Vals AI | – | 53.0 |
| Vals Finance Index | Vals AI | – | 58.0% |
| Vals RSI Index | Vals AI | – | 20.9% |
| Vals Multimodal Index | Vals AI | – | 73.4%† |
| Vals Router Index | Vals AI | – | 53.0%† |
| AA Cyber Index | Artificial Analysis | – | 41.3 |
| Safety & honesty | |||
| FORTRESS | Scale AI SEAL | – | 26.6% |
| Metadata | |||
| AA Openness Index | Artificial Analysis | – | 38.9 |
Data as of 2026-11-02. Best configuration of each model; every score links to its source on the model pages.
DuelLab GameBench 2, release gb2-20261026-cb69502a12d9 (https://benchmarks.duellab.org/releases/gb2-20261026-cb69502a12d9/), CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Scores mapped to BenchLeader model names; not endorsed by DuelLab.
What changed
- GLM-5 (thinking): cached input price moved from $0.15/M to $0.20/M (34% up).
- GLM-5 (thinking): cached input price moved from $0.20/M to $0.15/M (25% down).
- Kimi K3 (max): 1 new result (MirrorCode).
- Kimi K3 (max): 1 score revision.
- GLM-5 (thinking): BenchLeader Index 60.6 → 60.1.
- Kimi K3 (max): BenchLeader Index 63.1 → 62.1.
- GLM-5 (thinking): 1 new result (AA Progress Index v1); cached input price moved from $0.15/M to $0.20/M (34% up).
- Kimi K3 (max): 3 new results (AA Progress Index v1, HLE-Diamond (AA), CritPt v2 hard subset (AA)); 1 score revision.
- GLM-5 (thinking): cached input price moved from $0.20/M to $0.15/M (25% down).
- GLM-5 (thinking): cached input price moved from $0.15/M to $0.20/M (34% up).
Related comparisons
- GLM-5 vs Claude Opus 5.5
- GLM-5 vs Gemini 4 Carbon
- GLM-5 vs GPT-6 Astra
- Kimi K3 vs Claude Opus 5.5
- Kimi K3 vs Gemini 4 Carbon
- Kimi K3 vs GPT-6 Astra
- Kimi K3 vs GPT-6.1 Sol
- GLM-5 vs Claude Fable 5.1
Rankings: Best LLM overall · Best LLM for agents · Best cheap LLMs
GLM-5 vs Kimi K3: questions
- Is GLM-5 better than Kimi K3?
- GLM-5 (thinking) and Kimi K3 (max) are neck and neck on quality (60.2 vs 62.1, within the ±4.7 margin). The BenchLeader Index combines every independent quality benchmark; the two are within the margin of error as of 2 Nov 2026, so check the category scores for your use.
- GLM-5 vs Kimi K3: which is cheaper, and will that change?
- GLM-5 is cheaper today: $1.55 against $6.00 per million tokens, blended at three input tokens per output token. No price change has been announced for either.
- Is GLM-5 or Kimi K3 better for coding?
- Kimi K3 scores higher in coding (64 vs 59 on the category index, where 50 is average). On LiveCodeBench, the top shared coding benchmark, it is 81.9% for GLM-5 against 87.2% for Kimi K3.
- Which is faster, GLM-5 or Kimi K3?
- Kimi K3 streams faster (66 vs 62 output tokens per second, Artificial Analysis 7-day median).
- Is GLM-5 or Kimi K3 better for agents and tool use?
- Kimi K3 scores higher in agentic tasks (59 vs 58 on the category index, where 50 is average).
More questions (4)
- Which is better for long documents?
- Kimi K3 scores higher on long-context benchmarks (69 vs 63). Kimi K3 accepts more context: 1,048,576 against 204,800 tokens for GLM-5.
- Which has lower latency?
- Kimi K3 answers first: 37.2 s against 51.4 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.
- Do independent benchmarks agree on GLM-5 vs Kimi K3?
- Across the 11 benchmarks both have, GLM-5 leads 0, Kimi K3 leads 11. Every overall index that rates both puts Kimi K3 ahead.
- When was Kimi K3 released?
- Kimi K3 was released on 16 Jul 2026.