Kimi K2 Instruct: benchmark scores, price and ranking
Kimi K2 Instruct from Moonshot ranks #467 of 743 LLMs on the LLMs Tiger aggregate benchmark score (47.1), based on 7 benchmark results captured up to 2026-10-13.
- Organisation
- Moonshot
- Type
- Open-weight / local model
- Open weights
- Yes
- Released
- 2025-07-11 (source: curated)
- Cheapest API price
- $0.500 in / $2.00 out per million tokens (via deepinfra, observed 2026-10-13T20:47Z)
- Aggregate rank
- #467 of 743 (score 47.1)
Relative to other models, Kimi K2 Instruct is strongest on MMLU-Pro (#33 of 136) and weakest on SimpleBench (#49 of 61).
Across the 7 benchmarks where at least ten models are scored, Kimi K2 Instruct is in the top 10% on 0 and in the top half on 1.
Its cheapest listed input price, $0.500 per million tokens, is lower than that of 48% of the 427 models with a listed price.
Kimi K2 Instruct benchmark results
| Benchmark | Score | Rank | Source | Captured |
|---|---|---|---|---|
| MMLU-Pro | 81.0% | #33 of 136 | HuggingFace MMLU-Pro leaderboard | 2026-10-13 |
| GSO-Bench | 4.9% | #10 of 19 | Epoch AI Benchmarking Hub | 2026-10-13 |
| Fiction.liveBench | 61.1% | #26 of 47 | Epoch AI Benchmarking Hub | 2026-10-13 |
| WeirdML | 39.4% | #64 of 109 | Epoch AI Benchmarking Hub | 2026-10-13 |
| SWE-bench Verified | 43.8% | #48 of 73 | SWE-bench Verified leaderboard | 2026-10-13 |
| Terminal-Bench 2.0 | 27.8% | #20 of 28 | Epoch AI Benchmarking Hub | 2026-10-13 |
| SimpleBench | 26.3% | #49 of 61 | Epoch AI Benchmarking Hub | 2026-10-13 |
About Kimi K2 Instruct
Kimi K2 Instruct is an open-weight model that can run on your own hardware from Moonshot. It was released on 2025-07-11, according to the developer's announcement. The cheapest API listing we track charges $0.500 per million input tokens and $2.00 per million output tokens, through deepinfra. We hold 7 benchmark results for it across 3 areas, and none is in the top tenth of its board.
Where Kimi K2 Instruct stands
On our aggregate score, Kimi K2 Instruct is #467 of 743 ranked models, in the lower half. Its score is 47.1; the scale is centred on 50, so it sits 2.9 points below the typical model. That beats 37% of the ranked field. Among the 13 ranked Moonshot models it is #13. The next one up is Kimi K2 0905 (Novita) at 47.3.
Kimi K2 Instruct, benchmark by benchmark
MMLU-Pro puts Kimi K2 Instruct at 81.0%, #33 of 136. That beats 76% of the models tested, but 7.1 points short of Darwin-180B-RSI (88.1%). The results either side are JoyAI-LLM-Flash (81.0%) and qwen3-30b-a3b-thinking-2507 (80.9%). It covers 57-subject knowledge + reasoning. The board median is 55.3%, so Kimi K2 Instruct is 25.7 points above it. Source: HuggingFace MMLU-Pro leaderboard, captured 2026-10-13.
Kimi K2 Instruct lands below the middle on GSO-Bench: 4.9%, #10 of 19, with 47% of the field scoring higher. GPT-5.5 (xhigh) leads at 40.2%; the closest result above is claude-sonnet-4-20250514 at 4.9%. It covers software performance optimisation. The median score on this board is also 4.9%. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.
Kimi K2 Instruct lands below the middle on Fiction.liveBench: 61.1%, #26 of 47, with 53% of the field scoring higher. grok-4-0709 leads at 94.4%; the closest result above is Gemini 2.0 Flash (Feb 2025) at 61.1%. It covers long-context story comprehension (16k). The median score on this board is also 61.1%. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.
Kimi K2 Instruct lands below the middle on WeirdML: 39.4%, #64 of 109, with 58% of the field scoring higher. GPT-6 Astra (max) leads at 93.3%; the closest result above is GPT-4.5 Preview (Feb 2025) at 39.4%. It covers unusual machine-learning tasks. The board median is 41.2%, so Kimi K2 Instruct is 1.8 points below it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.
Kimi K2 Instruct lands below the middle on SWE-bench Verified: 43.8%, #48 of 73, with 64% of the field scoring higher. Claude Opus 4.7 (max) leads at 83.5%; the closest result above is Gemini 2.0 Flash (Experimental) at 44.2%. It covers real GitHub issues resolved. The board median is 56.4%, so Kimi K2 Instruct is 12.6 points below it. Source: SWE-bench Verified leaderboard, captured 2026-10-13.
Kimi K2 Instruct lands below the middle on Terminal-Bench 2.0: 27.8%, #20 of 28, with 68% of the field scoring higher. Gemini 3 Pro Preview leads at 69.4%; the closest result above is claude-haiku-4-5-20251001 at 29.8%. It covers real terminal tasks (agent + model). The board median is 37.3%, so Kimi K2 Instruct is 9.5 points below it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.
SimpleBench is a weak spot. Kimi K2 Instruct scores 26.3%, #49 of 61, behind 79% of the models measured. The leader, Claude Fable 5 (max), reaches 81.9%. It covers trick questions that test common sense. The board median is 41.4%, so Kimi K2 Instruct is 15.1 points below it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.
Strengths and weaknesses
Software engineering: uneven, ahead of 37% of the field on average across 3 benchmarks. Its best board here is GSO-Bench (#10 of 19) and its weakest is Terminal-Bench 2.0 (#20 of 28). Reasoning, maths and science: uneven, ahead of 48% of the field on average across 2 benchmarks. Its best board here is MMLU-Pro (#33 of 136) and its weakest is SimpleBench (#49 of 61). General knowledge and chat: uneven, ahead of 43% of the field on average across 2 benchmarks. Its best board here is Fiction.liveBench (#26 of 47) and its weakest is WeirdML (#64 of 109).
Price and value
The median input price across the 427 priced models is $0.500 per million tokens, so Kimi K2 Instruct costs about the same as the typical listing. Models that score within 1.5 points or better and cost less: Qwen2.5-Coder-7B-Instruct ($0.010, 49.5), DeepSeek V4.1 Flash (high) ($0.010, 48.9), Qwen2.5-Coder-3B-Instruct ($0.010, 46.5). Similar scores cost more on Kimi K2 Thinking ($0.550) and ernie-5.1 ($0.563).
Release history
The next one was Kimi K2 0905 (Novita) on 2025-09-05, at 47.3.
Reading these numbers
The 7 scores on this page come from 3 sources: Epoch AI Benchmarking Hub (5), HuggingFace MMLU-Pro leaderboard (1), SWE-bench Verified leaderboard (1). 1 of them come from an official leaderboard and 6 from datasets or third-party runs. With 7 results the aggregate rank can move a lot when one new score arrives.
Models ranked nearby
- #464 gpt-5.5-xhigh (codex-harness) (47.1)
- #465 AI21-Jamba-Large-1.5 (47.1)
- #466 GPT-5.4 nano (xhigh) (47.1)
- #468 Magistral Medium 1.2 (47.1)
- #469 Step 3.7 Flash (47.0)
- #470 DeepSeek-R1-Distill-Qwen-14B (47.0)
More from Moonshot
- Kimi K3 (max) (#15, 55.7)
- Kimi K2.5 (Fireworks) (#50, 52.8)
- Kimi K2.5 (#173, 50.7)
- Kimi K2 Thinking Turbo (#187, 50.5)
- Kimi K2 Thinking (Together) (#240, 49.7)
- Kimi-K2-Instruct-0905 (#258, 49.5)
Compare it on the live leaderboard → · How the aggregate score works