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 10 sources
As of 25 Oct 2026, Kimi K3 (max) and Muse Spark 1.2 (xhigh) are neck and neck on quality (64.8 vs 63.2 on the BenchLeader Index, inside the ±5.7 margin of error), and Muse Spark 1.2 (xhigh) costs 3.0× less, which makes it the better buy for most workloads. Pick Kimi K3 for coding (65 vs 55); pick Muse Spark 1.2 for knowledge (67 vs 63).
Kimi K3 benchmarks 
Muse Spark 1.2 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 | 65 vs 55 category score; Arena WebDev 1655 vs 1533 | High |
| Agents & tool use | Kimi K3 | 59 vs 55 category score; τ³-Banking (AA) 46.0% vs 34.9% | Medium |
| Maths | Muse Spark 1.2 | 67 vs 63 category score; Arena Maths 1491 vs 1499 | Medium |
| Reasoning | Muse Spark 1.2 | 69 vs 66 category score; CritPt 17.7% vs 23.4% | Medium |
| Knowledge | Muse Spark 1.2 | 67 vs 63 category score; SimpleQA Verified 60.3% vs 50.6% | Medium |
| Long documents | Kimi K3 | 70 vs 65 category score; AA-LCR 88.7% vs 79.0% | Medium |
| Images and other media | Muse Spark 1.2 | 65 vs 64 category score; SAGE 47.7% vs 52.8% | 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 | 17.6 s (Muse Spark 1.2 (xhigh)) vs 52.6 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 (64.8 vs 63.2, within the ±5.7 margin).
- Kimi K3 (max) is stronger in agents & tools, coding, composite, 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 7.6× faster (307 vs 41 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 | 66.4 | 69.3 | Muse Spark 1.2 |
| Coding | 64.6 | 55.1 | Kimi K3 |
| Agents & tools | 59.4 | 55.4 | Kimi K3 |
| Maths | 63.4 | 67.3 | Muse Spark 1.2 |
| Knowledge | 62.8 | 66.7 | Muse Spark 1.2 |
| Human preference | 68.1 | 68.6 | level |
| Multimodal | 63.6 | 65.2 | Muse Spark 1.2 |
| Long context | 69.6 | 64.5 | Kimi K3 |
| Composite | 78.3 | 73.8 | Kimi K3 |
Where sources disagree
Across the 50 benchmarks both have, Kimi K3 leads 37, Muse Spark 1.2 leads 13. Independent sources split: Kimi K3 leads on AA Intelligence Index v4.3.2; Muse Spark 1.2 leads on Arena Text.
| Overall index | Kimi K3 (max) | Muse Spark 1.2 (xhigh) | Leader |
|---|---|---|---|
| BenchLeader Index | 64.8 | 63.2 | neck and neck |
| AA Intelligence Index v4.3.2 | 43.6 | 39.6 | Kimi K3 |
| Arena Text | 1487 | 1491 | 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; suite cost is what Artificial Analysis paid to run its whole Intelligence Index.
| 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 |
| Cost to run a full benchmark suite | $2.00 | $0.975 |
| Output speed | 41 tok/s | 307 tok/s |
| Time to first answer | 52.6 s | 17.6 s |
| Time to a complete answer | 65.0 s | 19.2 s |
Per token is not per task: on Artificial Analysis' full suite Kimi K3 (max) cost $2.00 and Muse Spark 1.2 (xhigh) $0.975, because Muse Spark 1.2 (xhigh) used about 1.5× as many tokens for the same work, which a per-token price hides.
Each model is compared at its best-ranked reasoning-effort setting. Kimi K3 (max) scores 64.8 and takes 53 s to its first answer token; Kimi K3 scores 58.8 in 1 s, Kimi K3 (low) scores 58.2 in 57 s. Muse Spark 1.2 (xhigh) scores 63.2 and takes 18 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
134 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 | 64.8 | 63.2 |
| Agents & tools score | BenchLeader | 59.4 | 55.4 |
| Coding score | BenchLeader | 64.6 | 55.1 |
| Composite score | BenchLeader | 78.3 | 73.8 |
| Human preference score | BenchLeader | 68.1 | 68.6 |
| Knowledge score | BenchLeader | 62.8 | 66.7 |
| Long context score | BenchLeader | 69.6 | 64.5 |
| Maths score | BenchLeader | 63.4 | 67.3 |
| Multimodal score | BenchLeader | 63.6 | 65.2 |
| Reasoning score | BenchLeader | 66.4 | 69.3 |
| Blended price $/M | BenchLeader | $6.00 | $2.00 |
| Output speed | BenchLeader | 41 tok/s | 307 tok/s |
| Time to first answer | BenchLeader | 52.6 s | 17.6 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) | 1516 | 1512 |
| LiveBench 2.0 Reasoning | LiveBench | 90.7%† | 90.0%† |
| 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% |
| 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% |
| 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 | 64.6% | – |
| Coding | |||
| SciCode | SciCode | 59.5% | 57.4% |
| WeirdML | WeirdML | 82.6% | 60.3% |
| FrontierCode | Cognition | 44.2%† | – |
| Arena Coding | Arena (LMArena) | 1539 | 1541 |
| Arena WebDev | Arena (LMArena) | 1655 | 1533 |
| LiveBench 2.0 Coding | LiveBench | 59.3%† | 78.2%† |
| LiveBench 1.0 Coding | LiveBench | 81.5%† | 77.5% |
| LiveCodeBench | Vals AI | 87.2% | – |
| IOI | Vals AI | 48.9% | 21.8% |
Show all 134 benchmarks
| Metric | Publisher | Kimi K3 (max) | Muse Spark 1.2 (xhigh) |
|---|---|---|---|
| Coding | |||
| 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% |
| 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%† |
| FrontierCode 1.1 (AA) | Artificial Analysis | 43.6% | – |
| Agents & tools | |||
| APEX-Agents | Mercor | 47.8%† | 45.4%† |
| Arena Agent | Arena (LMArena) | 1.6 | -7 |
| LiveBench Long Agentic | LiveBench | 42.8%† | 21.3%† |
| LiveBench Browser | LiveBench | 16.0%† | 36.0%† |
| 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 | 32.0%† | – |
| 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 | 1499 | 1327 |
| 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 | 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) | 1499 | 1491 |
| Knowledge | |||
| SimpleQA Verified | Epoch AI Benchmarking Hub | 50.6% | 60.3% |
| LiveBench 2.0 Data Analysis | LiveBench | 78.7%† | 76.5%† |
| LiveBench Research | LiveBench | 79.0%† | 67.7%† |
| LiveBench 1.0 Data Analysis | LiveBench | 78.7%† | 76.5% |
| AA-Omniscience | Artificial Analysis | 19.7 | 27.2 |
| 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: accuracy | Artificial Analysis | 47.6% | 45.4% |
| AA-Omniscience: non-hallucination | Artificial Analysis | 46.8% | 66.7% |
| Instruction following | |||
| LiveBench Language | LiveBench | 85.5%† | 78.6% |
| LiveBench Instruction Following | LiveBench | 71.4%† | 74.3% |
| Arena Instruction Following | Arena (LMArena) | 1489 | 1472 |
| Human preference | |||
| Arena Text | Arena (LMArena) | 1487 | 1491 |
| EQ-Bench 4 | EQ-Bench | 1318† | – |
| Arena Creative Writing | Arena (LMArena) | 1462 | 1439 |
| Arena Multi-turn | Arena (LMArena) | 1502 | 1511 |
| Arena Longer Queries | Arena (LMArena) | 1503 | 1488 |
| Multimodal | |||
| Arena Vision | Arena (LMArena) | – | 1304 |
| LiveBench Design | LiveBench | 77.2%† | 67.1%† |
| 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 | 52.8% | 47.7% |
| Blueprint-Bench 2 | Epoch AI Benchmarking Hub | 29.5%† | – |
| AA-3D | Artificial Analysis | 982 | – |
| Long context | |||
| AA-LCR | Artificial Analysis | 88.7% | 79.0% |
| MLCR | Artificial Analysis | 21.7% | 7.2% |
| Composite | |||
| Epoch Capabilities Index | Epoch AI Benchmarking Hub | 157.9† | 155.1† |
| LiveBench 2.0 | LiveBench | 63.4%† | 62.4%† |
| LiveBench 1.0 | LiveBench | 79.2%† | 78.0% |
| AA Intelligence Index v4.3.2 | Artificial Analysis | 43.6 | 39.6 |
| 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 2026-10-25. Best configuration of each model; every score links to its source on the model pages.
What changed
- Muse Spark 1.2 (xhigh): 1 new result (AA-Briefcase v1.1).
- Kimi K3 (max): 7 new results (Code Migration v1.1, Excel Modeling Benchmark v1.1, Finance Agent v2.1…).
- Muse Spark 1.2 (xhigh): 1 new result (Harvey's Legal Agent Benchmark v1).
- Kimi K3 (max): 1 new result (Terminal-Bench-Science 0.1 (AA)); 1 score revision.
- Muse Spark 1.2 (xhigh): 2 score revisions.
- Kimi K3 (max): 3 score revisions.
- Muse Spark 1.2 (xhigh): BenchLeader Index 62.6 → 63.3.
- Kimi K3 (max): 1 score revision.
- Kimi K3 (max): 1 new result (τ³-bench Banking).
- Muse Spark 1.2 (xhigh): 4 new results (AutomationBench, GDP.pdf (AA), MLCR…).
Related comparisons
- Kimi K3 vs Claude Opus 5.5
- Kimi K3 vs Gemini 4 Argon
- Kimi K3 vs Muse Wonder Full
- Muse Spark 1.2 vs Claude Opus 5.5
- Muse Spark 1.2 vs Gemini 4 Argon
- Muse Spark 1.2 vs Muse Wonder Full
- Muse Spark 1.2 vs Claude Fable 5.1
- 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 (64.8 vs 63.2, within the ±5.7 margin). The BenchLeader Index combines every independent quality benchmark; the two are within the margin of error as of 2026-10-25, so check the category scores for your use.
- Which is cheaper, Kimi K3 or Muse Spark 1.2, 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 better than Muse Spark 1.2 for coding?
- Kimi K3 scores higher in coding (65 vs 55 on the category index, where 50 is average). On Arena WebDev, the top shared coding benchmark, it is 1655 for Kimi K3 against 1533 for Muse Spark 1.2.
- Is Kimi K3 better than Muse Spark 1.2 for agents and tool use?
- Kimi K3 scores higher in agentic tasks (59 vs 55 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.
- Which is better for long documents?
- Kimi K3 scores higher on long-context benchmarks (70 vs 65). Both accept 1,048,576 tokens of context.
- Which has lower latency?
- Muse Spark 1.2 (xhigh) answers first: 17.6 s against 52.6 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, Kimi K3 or Muse Spark 1.2?
- Muse Spark 1.2 streams faster: 307 against 41 output tokens per second.
- Do independent benchmarks agree on Kimi K3 vs Muse Spark 1.2?
- Across the 50 benchmarks both have, Kimi K3 leads 37, Muse Spark 1.2 leads 13. Independent sources split: Kimi K3 leads on AA Intelligence Index v4.3.2; Muse Spark 1.2 leads on Arena Text.
- When was Muse Spark 1.2 released?
- Muse Spark 1.2 was released on 5 Aug 2026.