Kimi K3 vs Muse Spark 1.2
Full profiles: Kimi K3 (Moonshot AI) and Muse Spark 1.2 (Meta): every benchmark, price and rank.
Updated · Data from 13 sources
As of 12 Nov 2026, Kimi K3 and Muse Spark 1.2 are neck and neck on quality (62.0 vs 61.6 on the BenchLeader Index, inside the ±5.9 margin), and Muse Spark 1.2 is 3.0× cheaper at $2.00 vs $6.00 per 1M tokens blended. Choose Kimi K3 for coding (64 vs 55), long context (69 vs 64), and agents & tools (58 vs 56); choose Muse Spark 1.2 for fast first answers (7.9 s vs 78.8 s) and reasoning (68 vs 59), knowledge (65 vs 60), and maths (67 vs 64), where it also streams 3×+ faster.
Key facts (as of 12 Nov 2026)
- Kimi K3 scores 62.0 on the BenchLeader Index and Muse Spark 1.2 scores 61.6 (BenchLeader, 12 Nov 2026).
- Kimi K3 costs $6.00 per 1M tokens blended (list price, 12 Nov 2026).
- Muse Spark 1.2 costs $2.00 per 1M tokens blended (list price, 12 Nov 2026).
- Muse Spark 1.2 streams 3×+ faster than Kimi K3 (Artificial Analysis, 7-day median, 12 Nov 2026).
- Of the 48 benchmarks both models have, Kimi K3 leads 36 (BenchLeader, 12 Nov 2026).
- On Arena WebDev, Kimi K3 is rated 1658 and Muse Spark 1.2 1534 (LMArena, as published by 12 Nov 2026).
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 55 category score; Arena WebDev 1658 vs 1534 | High |
| Agents & tool use | Kimi K3 | 58 vs 56 category score; τ³-Banking (AA) 46.0% vs 34.9% | Medium |
| Maths | Muse Spark 1.2 | 67 vs 64 category score; Arena Maths 1490 vs 1503 | Medium |
| Reasoning | Muse Spark 1.2 | 68 vs 59 category score; Arena Hard Prompts 1515 vs 1515 | High |
| Knowledge | Muse Spark 1.2 | 65 vs 60 category score; SimpleQA Verified 60.3% vs 50.6% | Medium |
| Long documents | Kimi K3 | 69 vs 64 category score; AA-LCR v1.1 88.7% vs 79.0% | Medium |
| Images and other media | Muse Spark 1.2 | 65 vs 63 category score | Medium |
| Chat quality (human preference) | Either | 68 vs 69 on Arena votes | Either |
| Lowest cost | Muse Spark 1.2 | $2.00 vs $6.00 per 1M blended; agentic coding session $0.092 vs $0.240 | – |
| Fastest first answer | Muse Spark 1.2 | 7.9 s (Muse Spark 1.2 (xhigh)) vs 78.8 s (Kimi K3 (max)) 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
- Kimi K3 (max) and Muse Spark 1.2 (xhigh) are neck and neck on quality (62.0 vs 61.6, within the ±5.9 margin).
- Kimi K3 (max) is stronger in agents & tools, coding, long context.
- Muse Spark 1.2 (xhigh) is stronger in human preference, knowledge, maths, multimodal, reasoning.
- Muse Spark 1.2 (xhigh) is 3.0× cheaper ($2.00 vs $6.00 per 1M blended).
- Muse Spark 1.2 (xhigh) streams 4.5× faster (204 vs 46 tokens per second).
At a glance
| Spec | Kimi K3 | Muse Spark 1.2 |
|---|---|---|
| Developer | Moonshot AI | Meta |
| Released | 16 Jul 2026 | 5 Aug 2026 |
| Context window (tokens) | 1,048,576 | 1,048,576 |
| Compared at | Kimi K3 (max) | Muse Spark 1.2 (xhigh) |
| Open weights | Yes | No |
| Price in / out / cached input per 1M | $3.00 / $15.00 / $0.300 | $1.25 / $4.25 / $0.150 |
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 | Kimi K3 (max) | Muse Spark 1.2 (xhigh) | Leader |
|---|---|---|---|
| Reasoning | 58.8 | 68.3 | Muse Spark 1.2 |
| Coding | 64.3 | 54.8 | Kimi K3 |
| Agents & tools | 58.3 | 55.5 | Kimi K3 |
| Maths | 64.2 | 67.2 | Muse Spark 1.2 |
| Knowledge | 59.6 | 64.5 | Muse Spark 1.2 |
| Human preference | 67.9 | 68.7 | level |
| Multimodal | 63.0 | 65.3 | Muse Spark 1.2 |
| Long context | 69.2 | 64.2 | 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.
Where sources disagree
Across the 48 benchmarks both have, Kimi K3 leads 36, Muse Spark 1.2 leads 12. Every overall index that rates both puts Muse Spark 1.2 ahead.
| Overall index | Kimi K3 (max) | Muse Spark 1.2 (xhigh) | Leader |
|---|---|---|---|
| BenchLeader Index | 62.0 | 61.6 | neck and neck |
| Arena Text | 1487 | 1494 | Muse Spark 1.2 |
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 | Kimi K3 (max) | Muse Spark 1.2 (xhigh) |
|---|---|---|
| Blended price per 1M tokens | $6.00 | $2.00 |
| Input / output per 1M | $3.00 / $15.00 | $1.25 / $4.25 |
| Cached input per 1M | $0.300 | $0.150 |
| Price per task: chat reply | $0.0057 | $0.0018 |
| Price per task: summarize a 30-page report | $0.045 | $0.018 |
| Price per task: code edit | $0.041 | $0.014 |
| Price per task: agentic coding session | $0.240 | $0.092 |
| Price per task: structured extraction | $0.0090 | $0.0034 |
| Agentic coding session, input mostly cached | $0.118 | $0.043 |
| Output speed | 46 tok/s | 204 tok/s |
| Time to first answer | 78.8 s | 7.9 s |
| Time to a complete answer | 156.1 s | 24.3 s |
Each model is compared at its best-ranked reasoning-effort setting. Kimi K3 (max) scores 62.0 and takes 79 s to its first answer token; Kimi K3 scores 59.2 in 1 s, Kimi K3 (low) scores 53.7 in 11 s. Muse Spark 1.2 (xhigh) scores 61.6 and takes 8 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
148 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 | Kimi K3 (max) | Muse Spark 1.2 (xhigh) |
|---|---|---|---|
| BenchLeader Index | BenchLeader | 62.0 | 61.6 |
| Agents & tools score | BenchLeader | 58.3 | 55.5 |
| Coding score | BenchLeader | 64.3 | 54.8 |
| Human preference score | BenchLeader | 67.9 | 68.7 |
| Knowledge score | BenchLeader | 59.6 | 64.5 |
| Long context score | BenchLeader | 69.2 | 64.2 |
| Maths score | BenchLeader | 64.2 | 67.2 |
| Multimodal score | BenchLeader | 63.0 | 65.3 |
| Reasoning score | BenchLeader | 58.8 | 68.3 |
| Blended price $/M | BenchLeader | $6.00 | $2.00 |
| Output speed | BenchLeader | 46 tok/s | 204 tok/s |
| Time to first answer | BenchLeader | 78.8 s | 7.9 s |
| Context window | BenchLeader | 1,048,576 tokens | 1,048,576 tokens |
| Reasoning | |||
| GPQA Diamond | Epoch AI Benchmarking Hub | 93.1% | – |
| SimpleBench | SimpleBench | 60.7% | 74.5%† |
| Arena Hard Prompts | Arena (LMArena) | 1515 | 1515 |
| LiveBench 2.0 Reasoning | LiveBench | 90.7%† | – |
| LiveBench 1.0 Reasoning | LiveBench | 90.7%† | 90.0% |
| GPQA Diamond (AA) | Artificial Analysis | 93.5% | 90.4% |
| Humanity's Last Exam (AA) | Artificial Analysis | 46.9% | 45.5% |
| HLE-Diamond (AA) | Artificial Analysis | 26.9% | – |
| GPQA Diamond (Vals) | Vals AI | 92.9%† | – |
| ARC-AGI-1 | ARC Prize | 94.5% | – |
| ARC-AGI-2 | ARC Prize | 60.4% | – |
| CritPt | Artificial Analysis | 23.4% | 17.7% |
| CritPt v2 hard subset (AA) | Artificial Analysis | 40.5% | – |
| Chess Puzzles | Epoch AI Benchmarking Hub | 36.0% | 27.0% |
| Mystery Game Puzzles | Epoch AI Benchmarking Hub | 26.0% | – |
| LMCA | Epoch AI Benchmarking Hub | 52.7% | 48.4% |
| DTBench | Epoch AI Benchmarking Hub | 91.2% | 94.7% |
| ForecastBench | ForecastBench | 64.2% | – |
| Coding | |||
| SciCode | SciCode | 59.5% | 57.4% |
| WeirdML | WeirdML | 82.6% | 60.3% |
| FrontierCode | Cognition | 44.2%† | – |
| Arena Coding | Arena (LMArena) | 1537 | 1543 |
| Arena WebDev | Arena (LMArena) | 1658 | 1534 |
| LiveBench 2.0 Coding | LiveBench | 53.1%† | – |
| LiveBench 1.0 Coding | LiveBench | 81.5%† | 77.5% |
Show all 148 benchmarks
| Metric | Publisher | Kimi K3 (max) | Muse Spark 1.2 (xhigh) |
|---|---|---|---|
| Coding | |||
| LiveCodeBench | Vals AI | 87.2% | – |
| IOI | Vals AI | 48.9% | 21.8% |
| SWE-bench (Vals) | Vals AI | 93.4%† | 86.6% |
| Code Migration | Vals AI | 16.1% | 29.9% |
| Code Migration v1.1 | Vals AI | 28.6% | – |
| ProgramBench | Vals AI | 2.0%† | 0.5% |
| SWE-bench Multimodal | Vals AI | 44.2% | – |
| Vibe Code Bench 1-100 | Vals AI | 18.2% | – |
| Vibe Code Bench v1.1 | Vals AI | 85.0% | 79.1% |
| MirrorCode | Epoch AI Benchmarking Hub | 54.4% | – |
| Surface Evolver Bench | Epoch AI Benchmarking Hub | 93.0% | – |
| DeepSWE v1.1 | Epoch AI Benchmarking Hub | 68.5% | 54.9% |
| ALE-Bench | Epoch AI Benchmarking Hub | 1524.5 | – |
| FrontierSWE | Epoch AI Benchmarking Hub | 25.9% | 12.0% |
| Terminal-Bench 4.0 (AA) | Artificial Analysis | 12.6% | 7.1% |
| Terminal-Bench 2.1 (AA) | Artificial Analysis | 85.0% | 80.2% |
| Startrise Frontend | Startrise | 79.8%† | 70.3%† |
| GameBench 2 Overall | DuelLab GameBench 2 | 47.6† | 51.4† |
| GameBench 2 by reasoning setting | DuelLab GameBench 2 | 44.8 | 57.3 |
| Design Arena Website | Design Arena | 1329† | 1309† |
| FrontierCode 1.1 (AA) | Artificial Analysis | 43.6% | – |
| FrontierCode 2.0 (AA) | Artificial Analysis | 10.9%† | – |
| Agents & tools | |||
| APEX-Agents | Mercor | 47.8%† | 45.4%† |
| Arena Agent | Arena (LMArena) | 0.6 | -8.2 |
| LiveBench Long Agentic | LiveBench | 36.0%† | – |
| LiveBench Browser | LiveBench | 75.2%† | – |
| LiveBench Agentic Coding | LiveBench | 62.2%† | 57.6% |
| Terminal-Bench 2.1 (Vals) | Vals AI | 80.9%† | 69.7% |
| MCP Atlas | Scale AI SEAL | 82.3% | – |
| HiL-Bench | Scale AI SEAL | 42.7%† | – |
| τ³-bench Banking | τ²-bench | 37.1% | – |
| τ³-Banking (AA) | Artificial Analysis | 46.0% | 34.9% |
| ITBench SRE (AA) | Artificial Analysis | 47.7% | – |
| Analyst Agent (AA) | Artificial Analysis | 38.8% | – |
| APEX-Agents (AA) | Artificial Analysis | 41.3% | – |
| CyberBench | Vals AI | 75.2% | 69.5% |
| Finance Agent v2 | Vals AI | 53.1% | 60.6% |
| 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% | 25.4% |
| Legal Research Bench | Vals AI | 46.1% | 43.8% |
| SkillsBench | Vals AI | – | 53.0% |
| SREBench | Vals AI | – | 0.0% |
| Tax Agent Bench | Vals AI | 33.4% | 32.8% |
| Terminal-Bench 4.0 (Vals) | Vals AI | 17.2% | 6.1% |
| Terminal-Bench Science | Vals AI | 1.4% | – |
| Time Horizon Index: KSP | Vals AI | 10.5% | – |
| PostTrainBench | Epoch AI Benchmarking Hub | 31.5%† | – |
| GBAEval | Epoch AI Benchmarking Hub | 48.4%† | – |
| Vending-Bench 2 | Andon Labs | 5165.0† | – |
| AutomationBench | Artificial Analysis | 58.3% | 40.6% |
| GDP.pdf (AA) | Artificial Analysis | 22.0% | 17.4% |
| Harvey LAB-AA v1.1 | Harvey | 5.3% | – |
| EnterpriseOps-Gym | Artificial Analysis | 45.3% | 47.3% |
| AA-Briefcase v1.1 | Artificial Analysis | 1497 | 1324 |
| GDPx | GDPx by SemiAnalysis | 3.7% | – |
| Design Arena Agentic | Design Arena | 1255† | 1179† |
| 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% | – |
| FACTS Search | 90.3%† | – | |
| 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 | 97.2% | – |
| ProofBench | Vals AI | 87.0%† | 43.0%† |
| LiveBench Mathematics | LiveBench | 84.4%† | 91.2% |
| AIME 2026 | MathArena | 96.7% | – |
| HMMT February 2026 | MathArena | 97.0% | – |
| MathArena Apex | MathArena | 65.6% | – |
| Arena Maths | Arena (LMArena) | 1503 | 1490 |
| Knowledge | |||
| SimpleQA Verified | Epoch AI Benchmarking Hub | 50.6% | 60.3% |
| LiveBench 2.0 Data Analysis | LiveBench | 78.7%† | – |
| LiveBench Research | LiveBench | 79.8%† | – |
| LiveBench 1.0 Data Analysis | LiveBench | 78.7%† | 76.5% |
| AA-Omniscience v1.1 | Artificial Analysis | -1.7† | – |
| MMLU-Pro | Vals AI | 88.0% | 88.3% |
| LegalBench | Vals AI | 86.2% | 85.3% |
| CorpFin | Vals AI | 71.6%† | 70.9% |
| TaxEval | Vals AI | 75.7%† | 80.4% |
| BioMysteryBench | Vals AI | 72.6% | 64.8% |
| Excel Modeling Benchmark | Vals AI | 66.7% | 57.0% |
| Excel Modeling Benchmark v1.1 | Vals AI | 71.3% | – |
| MedCode | Vals AI | 49.4% | 49.4% |
| MedScribe | Vals AI | 88.0% | 90.1% |
| Public Benefits Bench v1.1 | Vals AI | 68.2% | 68.5% |
| GDP.pdf | Epoch AI Benchmarking Hub | 19.0% | 16.0% |
| 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%† | 78.6% |
| LiveBench Instruction Following | LiveBench | 71.4%† | 74.3% |
| Arena Instruction Following | Arena (LMArena) | 1488 | 1479 |
| Human preference | |||
| Arena Text | Arena (LMArena) | 1487 | 1494 |
| EQ-Bench 4 | EQ-Bench | 1319† | – |
| Arena Creative Writing | Arena (LMArena) | 1459 | 1443 |
| Arena Multi-turn | Arena (LMArena) | 1501 | 1512 |
| Arena Longer Queries | Arena (LMArena) | 1500 | 1491 |
| Multimodal | |||
| Arena Vision | Arena (LMArena) | – | 1305 |
| LiveBench Design | LiveBench | 72.5%† | – |
| MMMU-Pro | Artificial Analysis | 80.5% | – |
| MMMU-Pro (Vals) | Vals AI | 88.2%† | 86.1% |
| MortgageTax | Vals AI | 66.3%† | 65.4% |
| SAGE | Vals AI | 86.8% | – |
| Blueprint-Bench 2 | Epoch AI Benchmarking Hub | 29.5%† | – |
| Code2Video-AA | Artificial Analysis | 1041 | – |
| AA-3D | Artificial Analysis | 984 | – |
| Long context | |||
| AA-LCR v1.1 | Artificial Analysis | 88.7% | 79.0% |
| MLCR | Artificial Analysis | 21.7% | 7.2% |
| Composite | |||
| Epoch Capabilities Index | Epoch AI Benchmarking Hub | 158.6† | 155.0† |
| LiveBench 2.0 | LiveBench | 65.0%† | – |
| LiveBench 1.0 | LiveBench | 79.2%† | 78.0% |
| AA Intelligence Index v5.0 | Artificial Analysis | 17.4† | – |
| AA Intelligence Index v4.3.2 | Artificial Analysis | 43.6 | 39.6 |
| AA Progress Index v1 | Artificial Analysis | 319.9 | 309.1 |
| 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%† | 69.8%† |
| 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 2026-11-12. Best configuration of each model; every score links to its source on the model pages.
DuelLab GameBench 2, release gb2-20261110-2acac640a228 (https://benchmarks.duellab.org/releases/gb2-20261110-2acac640a228/), CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Scores mapped to BenchLeader model names; not endorsed by DuelLab.
What changed
- Kimi K3 (max): 3 new results (AIME 2026, HMMT February 2026, MathArena Apex).
- Kimi K3 (max): 1 new result (MirrorCode); 1 score revision.
- Kimi K3 (max): 1 score revision.
- Kimi K3 (max): 2 score revisions.
- Muse Spark 1.2 (xhigh): 1 score revision.
- Kimi K3 (max): 1 new result (MirrorCode).
- Kimi K3 (max): 1 score revision.
- Muse Spark 1.2 (xhigh): 1 score revision.
- Kimi K3 (max): BenchLeader Index 63.1 → 62.1.
- Muse Spark 1.2 (xhigh): BenchLeader Index 62.2 → 61.6.
Related comparisons
- Kimi K3 vs Claude Fable 5.2
- Kimi K3 vs Gemini 4 Carbon
- Kimi K3 vs Claude Opus 5.5
- Muse Spark 1.2 vs Claude Fable 5.2
- Muse Spark 1.2 vs Gemini 4 Carbon
- Muse Spark 1.2 vs Claude Opus 5.5
- Muse Spark 1.2 vs GPT-6 Astra
- Kimi K3 vs Muse Spark 1.1
Rankings: Best LLM overall · Best LLM for agents · Best cheap LLMs
Kimi K3 vs Muse Spark 1.2: questions
- Is Kimi K3 better than Muse Spark 1.2?
- Kimi K3 (max) and Muse Spark 1.2 (xhigh) are neck and neck on quality (62.0 vs 61.6, within the ±5.9 margin). The BenchLeader Index combines every independent quality benchmark; the two are within the margin of error as of 12 Nov 2026, so check the category scores for your use.
- Kimi K3 vs Muse Spark 1.2: which is cheaper, and will that change?
- Muse Spark 1.2 is cheaper today: $2.00 against $6.00 per million tokens, blended at three input tokens per output token. No price change has been announced for either.
- Is Kimi K3 or Muse Spark 1.2 better for coding?
- Kimi K3 scores higher in coding (64 vs 55 on the category index, where 50 is average). On Arena WebDev, the top shared coding benchmark, it is 1658 for Kimi K3 against 1534 for Muse Spark 1.2.
- Which is faster, Kimi K3 or Muse Spark 1.2?
- Muse Spark 1.2 streams 3×+ faster (204 vs 46 output tokens per second, Artificial Analysis 7-day median).
- Is Kimi K3 or Muse Spark 1.2 better for agents and tool use?
- Kimi K3 scores higher in agentic tasks (58 vs 56 on the category index, where 50 is average). On τ³-Banking (AA), the top shared agents & tools benchmark, it is 46.0% for Kimi K3 against 34.9% for Muse Spark 1.2.
More questions (4)
- Which is better for long documents?
- Kimi K3 scores higher on long-context benchmarks (69 vs 64). Both accept 1,048,576 tokens of context.
- Which has lower latency?
- Muse Spark 1.2 answers first: 7.9 s against 78.8 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 Kimi K3 vs Muse Spark 1.2?
- Across the 48 benchmarks both have, Kimi K3 leads 36, Muse Spark 1.2 leads 12. Every overall index that rates both puts Muse Spark 1.2 ahead.
- When was Muse Spark 1.2 released?
- Muse Spark 1.2 was released on 5 Aug 2026.