Kimi K2 Thinking vs MiniMax M3.1 Flash
Full profiles: Kimi K2 Thinking (Moonshot AI) and MiniMax M3.1 Flash (MiniMax): every benchmark, price and rank.
Updated · Data from 12 sources
As of 16 Nov 2026, Kimi K2 Thinking and MiniMax M3.1 Flash are neck and neck on quality (52.3 vs 54.3 on the BenchLeader Index, inside the ±5.6 margin), and MiniMax M3.1 Flash is 4.1× cheaper at $0.262 vs $1.07 per 1M tokens blended. Choose Kimi K2 Thinking for knowledge (51 vs 46) and fast first answers (25.1 s vs 43.3 s); choose MiniMax M3.1 Flash for coding (55 vs 49), long context (65 vs 61), and agents & tools (55 vs 51).
Key facts (as of 16 Nov 2026)
- Kimi K2 Thinking scores 52.3 on the BenchLeader Index and MiniMax M3.1 Flash scores 54.3 (BenchLeader, 16 Nov 2026).
- Kimi K2 Thinking costs $1.07 per 1M tokens blended (list price, 16 Nov 2026).
- MiniMax M3.1 Flash costs $0.262 per 1M tokens blended (list price, 16 Nov 2026).
- Of the 8 benchmarks both models have, MiniMax M3.1 Flash leads 7 (BenchLeader, 16 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 | MiniMax M3.1 Flash | 55 vs 49 category score; Design Arena Website 1277 vs 1105 | Medium |
| Agents & tool use | MiniMax M3.1 Flash | 55 vs 51 category score | Medium |
| Reasoning | MiniMax M3.1 Flash | 50 vs 47 category score; Humanity's Last Exam (AA) 48.0% vs 23.8% | Medium |
| Knowledge | Kimi K2 Thinking | 51 vs 46 category score; PRBench Finance 43.4% vs 45.6% | Medium |
| Long documents | MiniMax M3.1 Flash | 65 vs 61 category score; AA-LCR v1.1 79.7% vs 72.0% | Medium |
| Lowest cost | MiniMax M3.1 Flash | $0.262 vs $1.07 per 1M blended; agentic coding session $0.011 vs $0.046 | – |
| Fastest first answer | Kimi K2 Thinking | 25.1 s (Kimi K2 Thinking (thinking)) vs 43.3 s (MiniMax M3.1 Flash) 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 K2 Thinking (thinking) and MiniMax M3.1 Flash are neck and neck on quality (52.3 vs 54.3, within the ±5.6 margin).
- Kimi K2 Thinking (thinking) is stronger in knowledge.
- MiniMax M3.1 Flash is stronger in agents & tools, coding, long context, reasoning.
- MiniMax M3.1 Flash is 4.1× cheaper ($0.262 vs $1.07 per 1M blended).
At a glance
| Spec | Kimi K2 Thinking | MiniMax M3.1 Flash |
|---|---|---|
| Developer | Moonshot AI | MiniMax |
| Released | 6 Nov 2025 | 19 Oct 2026 |
| Context window (tokens) | 262,144 | 1,000,000 |
| Compared at | Kimi K2 Thinking (thinking) | MiniMax M3.1 Flash |
| Open weights | Yes | Yes |
| Price in / out / cached input per 1M | $0.600 / $2.50 / $0.150 | $0.150 / $0.600 / $0.015 |
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 K2 Thinking (thinking) | MiniMax M3.1 Flash | Leader |
|---|---|---|---|
| Reasoning | 46.9 | 50.1 | MiniMax M3.1 Flash |
| Coding | 49.2 | 55.4 | MiniMax M3.1 Flash |
| Agents & tools | 51.4 | 55.2 | MiniMax M3.1 Flash |
| Maths | 50.5 | – | not comparable |
| Knowledge | 51.0 | 45.9 | Kimi K2 Thinking |
| Instruction following | 59.3 | – | not comparable |
| Multimodal | – | 61.5 | not comparable |
| Long context | 60.5 | 64.5 | MiniMax M3.1 Flash |
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: Maths, Instruction following, Multimodal, Human preference.
Where sources disagree
Across the 8 benchmarks both have, Kimi K2 Thinking leads 0, MiniMax M3.1 Flash leads 7 and 1 is a tie.
| Overall index | Kimi K2 Thinking (thinking) | MiniMax M3.1 Flash | Leader |
|---|---|---|---|
| BenchLeader Index | 52.3 | 54.3 | neck and neck |
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 K2 Thinking (thinking) | MiniMax M3.1 Flash |
|---|---|---|
| Blended price per 1M tokens | $1.07 | $0.262 |
| Input / output per 1M | $0.600 / $2.50 | $0.150 / $0.600 |
| Cached input per 1M | $0.150 | $0.015 |
| Price per task: chat reply | $0.0010 | $0.0002 |
| Price per task: summarize a 30-page report | $0.0087 | $0.0022 |
| Price per task: code edit | $0.0074 | $0.0018 |
| Price per task: agentic coding session | $0.046 | $0.011 |
| Price per task: structured extraction | $0.0017 | $0.0004 |
| Agentic coding session, input mostly cached | $0.026 | $0.0053 |
| Cost per benchmark task (AA Index) | – | $0.491 |
| Output tokens per task (AA Index) | – | 138k |
| Of which reasoning | – | 103k |
| Time per task (AA Index) | – | 15 min |
| Output speed | 130 tok/s | 151 tok/s |
| Time to first answer | 25.1 s | 43.3 s |
| Time to a complete answer, last measured 13 Nov 2026 | 46.4 s | 141.5 s |
Each model is compared at its best-ranked reasoning-effort setting. Kimi K2 Thinking (thinking) scores 52.3 and takes 25 s to its first answer token. MiniMax M3.1 Flash scores 54.3 and takes 43 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
94 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 K2 Thinking (thinking) | MiniMax M3.1 Flash |
|---|---|---|---|
| BenchLeader Index | BenchLeader | 52.3 | 54.3 |
| Agents & tools score | BenchLeader | 51.4 | 55.2 |
| Coding score | BenchLeader | 49.2 | 55.4 |
| Instruction following score | BenchLeader | 59.3 | – |
| Knowledge score | BenchLeader | 51.0 | 45.9 |
| Long context score | BenchLeader | 60.5 | 64.5 |
| Maths score | BenchLeader | 50.5 | – |
| Reasoning score | BenchLeader | 46.9 | 50.1 |
| Multimodal score | BenchLeader | – | 61.5 |
| Blended price $/M | BenchLeader | $1.07 | $0.262 |
| Output speed | BenchLeader | 130 tok/s | 151 tok/s |
| Time to first answer | BenchLeader | 25.1 s | 43.3 s |
| Context window | BenchLeader | 262,144 tokens | 1,000,000 tokens |
| Reasoning | |||
| SimpleBench | SimpleBench | 39.6% | – |
| LiveBench 2.0 Reasoning | LiveBench | – | 82.9%† |
| GPQA Diamond (AA) | Artificial Analysis | 83.8% | – |
| Humanity's Last Exam (AA) | Artificial Analysis | 23.8% | 48.0% |
| HLE-Diamond (AA) | Artificial Analysis | – | 28.1% |
| GPQA Diamond (Vals) | Vals AI | 78.5% | – |
| Kagi LLM Benchmark | Kagi LLM Benchmark | 64.4% | – |
| CritPt | Artificial Analysis | 2.6% | 22.3% |
| CritPt v2 hard subset (AA) | Artificial Analysis | – | 32.5% |
| CL-bench | Epoch AI Benchmarking Hub | 17.6% | – |
| Coding | |||
| WeirdML | WeirdML | 42.8% | – |
| Arena WebDev | Arena (LMArena) | – | 1584 |
| LiveBench 2.0 Coding | LiveBench | – | 68.8%† |
| SciCode (AA) | Artificial Analysis | – | 53.4% |
| LiveCodeBench | Vals AI | 63.1% | – |
| SWE-bench (Vals) | Vals AI | 60.2% | – |
| SWE-bench Verified (bash only) | SWE-bench | 63.4% | – |
| SWE-bench Verified (any scaffold) | SWE-bench | 63.4% | – |
| Vibe Code Bench v1.1 | Vals AI | – | 73.9%† |
| Vibe Code Bench 1-100 v1.1 | Vals AI | – | 19.8%† |
| SWE-Bench Pro V2 | Scale AI SEAL | – | 96.6%† |
| SWE-Bench Pro V2 HARD | Scale AI SEAL | – | 84.3%† |
| ALE-Bench | Epoch AI Benchmarking Hub | 597.5 | – |
| Terminal-Bench 4.0 (AA) | Artificial Analysis | – | 25.8% |
| GameBench 2 by reasoning setting | DuelLab GameBench 2 | – | 30.3† |
Show all 94 benchmarks
| Metric | Publisher | Kimi K2 Thinking (thinking) | MiniMax M3.1 Flash |
|---|---|---|---|
| Coding | |||
| Design Arena Website | Design Arena | 1105 | 1277 |
| FrontierCode 1.1 (AA) | Artificial Analysis | – | 38.6% |
| FrontierCode 2.0 (AA) | Artificial Analysis | – | 20.8% |
| LiveCodeBench (AA) | Artificial Analysis | 85.3% | – |
| Agents & tools | |||
| Terminal-Bench | Terminal-Bench | 35.7% | – |
| Arena Agent | Arena (LMArena) | – | -2.7 |
| LiveBench Long Agentic | LiveBench | – | 25.7%† |
| LiveBench Browser | LiveBench | – | 80.6%† |
| Terminal-Bench Hard | Artificial Analysis | 31.1% | – |
| τ²-Bench Telecom (AA) | Artificial Analysis | 93.0% | – |
| HiL-Bench | Scale AI SEAL | – | 65.7% |
| GDPval-AA v2.1, normalised score | Artificial Analysis | – | 47.9% |
| GDPval-AA v2.1 | Artificial Analysis | – | 1465 |
| Finance Agent v2.1 | Vals AI | – | 58.8%† |
| Harvey's Legal Agent Benchmark v1.1 | Vals AI | – | 7.5%† |
| Legal Research Bench | Vals AI | – | 34.1%† |
| Poker Agent | Vals AI | 1011.6% | – |
| Tax Agent Bench | Vals AI | – | 32.1%† |
| Terminal-Bench 2.0 (Vals) | Vals AI | 37.1% | – |
| Terminal-Bench 4.0 (Vals) | Vals AI | – | 25.3%† |
| Arena Agent Task Completion | Arena (LMArena) | – | -0.7 |
| METR Time Horizons | METR | 59.2% | – |
| AutomationBench | Artificial Analysis | – | 72.5% |
| GDP.pdf (AA) | Artificial Analysis | – | 20.8% |
| AA-Briefcase v1.1 | Artificial Analysis | – | 1474 |
| GDPx | GDPx by SemiAnalysis | – | 4.0%† |
| Design Arena Agentic | Design Arena | – | 1252 |
| Terminal-Bench-Science 0.1 (AA) | Artificial Analysis | – | 3.3% |
| MKA (AA) | Artificial Analysis | – | 33.5% |
| Maths | |||
| AIME (Vals) | Vals AI | 85.4% | – |
| MGSM | Vals AI | 90.2% | – |
| MathArena Apex | MathArena | 0.0% | – |
| AIME 2025 (AA) | Artificial Analysis | 94.7% | – |
| Knowledge | |||
| LiveBench 2.0 Data Analysis | LiveBench | – | 71.4%† |
| LiveBench Research | LiveBench | – | 76.7%† |
| AA-Omniscience v1.1 | Artificial Analysis | – | -6.4 |
| MMLU-Pro | Vals AI | 81.1% | – |
| LegalBench | Vals AI | 80.2% | – |
| CorpFin | Vals AI | 60.6% | – |
| TaxEval | Vals AI | 71.7% | – |
| MedQA | Vals AI | 92.6% | – |
| PRBench Finance | Scale AI SEAL | 43.4% | 45.6% |
| PRBench Legal | Scale AI SEAL | 40.9% | 45.1% |
| CaseLaw v2 | Vals AI | 65.7% | – |
| Excel Modeling Benchmark v1.1 | Vals AI | – | 67.8%† |
| Access to Justice Benchmark | Vals AI | – | 73.3%† |
| AA-Omniscience v1.1: accuracy | Artificial Analysis | – | 45.1% |
| AA-Omniscience v1.1: non-hallucination | Artificial Analysis | – | 6.3% |
| Global-MMLU-Lite (AA) | Artificial Analysis | 73.5% | – |
| Instruction following | |||
| IFBench | Artificial Analysis | 68.1% | – |
| MultiChallenge | Scale AI SEAL | 55.4% | – |
| Human preference | |||
| EQ-Bench Creative Writing v3 | EQ-Bench | 1631 | – |
| Multimodal | |||
| LiveBench Design | LiveBench | – | 71.2%† |
| MMMU-Pro | Artificial Analysis | – | 79.1% |
| SAGE | Vals AI | – | 85.3%† |
| VIALS-AA | Handshake AI | – | 29.8% |
| Long context | |||
| AA-LCR v1.1 | Artificial Analysis | 72.0% | 79.7% |
| MLCR | Artificial Analysis | – | 16.7% |
| Composite | |||
| Epoch Capabilities Index | Epoch AI Benchmarking Hub | 145.4 | – |
| LiveBench 2.0 | LiveBench | – | 63.7%† |
| AA Intelligence Index v5.0.1 | Artificial Analysis | – | 29.1 |
| AA Intelligence Index v4.3.2 | Artificial Analysis | 22.0 | 42.7 |
| AA Progress Index v1 | Artificial Analysis | 264.7 | 317.6 |
| Vals Index | Vals AI | – | 53.8† |
| Vals Finance Index | Vals AI | – | 55.9%† |
| Vals Router Index | Vals AI | – | 53.8% |
| Vals Legal Index | Vals AI | – | 67.1%† |
| Safety & honesty | |||
| FORTRESS | Scale AI SEAL | – | 52.1% |
| Metadata | |||
| AA Openness Index | Artificial Analysis | 27.8 | 27.8 |
Data as of 2026-11-16. Best configuration of each model; every score links to its source on the model pages.
DuelLab GameBench 2, release gb2-20261111-7b598bb13de4 (https://benchmarks.duellab.org/releases/gb2-20261111-7b598bb13de4/), CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Scores mapped to BenchLeader model names; not endorsed by DuelLab.
What changed
- Kimi K2 Thinking (thinking): BenchLeader Index 51.8 → 52.3.
- MiniMax M3.1 Flash: BenchLeader Index 53.7 → 54.3.
- Kimi K2 Thinking (thinking): 2 new results (Global-MMLU-Lite (AA), AIME 2025 (AA)).
- Kimi K2 Thinking (thinking): BenchLeader Index 53.1 → 51.8.
- MiniMax M3.1 Flash: BenchLeader Index 54.3 → 53.7.
- Kimi K2 Thinking (thinking): 1 new result (SimpleBench).
- MiniMax M3.1 Flash: 1 score revision.
- MiniMax M3.1 Flash: 6 score revisions.
- MiniMax M3.1 Flash: BenchLeader Index 53.0 → 54.3.
- MiniMax M3.1 Flash: 1 new result (HiL-Bench).
Related comparisons
- MiniMax M3.1 Flash vs Claude Opus 5.5
- MiniMax M3.1 Flash vs GPT-6 Astra
- MiniMax M3.1 Flash vs Claude Fable 5.1
- MiniMax M3.1 Flash vs GPT-6.1 Sol
- MiniMax M3.1 Flash vs Gemini 4 Argon
- MiniMax M3.1 Flash vs GPT-5.6 Sol
- MiniMax M3.1 Flash vs Claude Sonnet 5.5
- MiniMax M3.1 Flash vs Muse Wonder Full
Rankings: Best LLM overall · Best LLM for agents · Best cheap LLMs
Kimi K2 Thinking vs MiniMax M3.1 Flash: questions
- Is Kimi K2 Thinking better than MiniMax M3.1 Flash?
- Kimi K2 Thinking (thinking) and MiniMax M3.1 Flash are neck and neck on quality (52.3 vs 54.3, within the ±5.6 margin). The BenchLeader Index combines every independent quality benchmark; the two are within the margin of error as of 16 Nov 2026, so check the category scores for your use.
- Kimi K2 Thinking vs MiniMax M3.1 Flash: which is cheaper, and will that change?
- MiniMax M3.1 Flash is cheaper today: $0.262 against $1.07 per million tokens, blended at three input tokens per output token. No price change has been announced for either.
- Is Kimi K2 Thinking or MiniMax M3.1 Flash better for coding?
- MiniMax M3.1 Flash scores higher in coding (55 vs 49 on the category index, where 50 is average).
- Which is faster, Kimi K2 Thinking or MiniMax M3.1 Flash?
- MiniMax M3.1 Flash streams faster (151 vs 130 output tokens per second, Artificial Analysis 7-day median).
- Is Kimi K2 Thinking or MiniMax M3.1 Flash better for agents and tool use?
- MiniMax M3.1 Flash scores higher in agentic tasks (55 vs 51 on the category index, where 50 is average).
More questions (4)
- Which is better for long documents?
- MiniMax M3.1 Flash scores higher on long-context benchmarks (65 vs 61). MiniMax M3.1 Flash accepts more context: 1,000,000 against 262,144 tokens for Kimi K2 Thinking.
- Which has lower latency?
- Kimi K2 Thinking answers first: 25.1 s against 43.3 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 K2 Thinking vs MiniMax M3.1 Flash?
- Across the 8 benchmarks both have, Kimi K2 Thinking leads 0, MiniMax M3.1 Flash leads 7 and 1 is a tie.
- When was MiniMax M3.1 Flash released?
- MiniMax M3.1 Flash was released on 19 Oct 2026.