LLMs TigerLive LLM benchmarks

Kimi K2.5: benchmark scores, price and ranking

Kimi K2.5 from Moonshot ranks #173 of 743 LLMs on the LLMs Tiger aggregate benchmark score (50.7), based on 21 benchmark results captured up to 2026-10-13.

Organisation
Moonshot
Type
Open-weight / local model
Open weights
Yes
Released
2026-01-27 (source: epoch)
Cheapest API price
$0.450 in / $2.25 out per million tokens (via deepinfra, observed 2026-10-13T19:47Z)
Aggregate rank
#173 of 743 (score 50.7)

Relative to other models, Kimi K2.5 is strongest on MMLU-Pro (#5 of 136) and weakest on SWE-bench Multilingual (#7 of 11).

Across the 19 benchmarks where at least ten models are scored, Kimi K2.5 is in the top 10% on 3 and in the top half on 15.

Its cheapest listed input price, $0.450 per million tokens, is lower than that of 51% of the 427 models with a listed price.

Kimi K2.5 benchmark results

Kimi K2.5 benchmark scores, ranked against all models measured on each benchmark
BenchmarkScoreRankSourceCaptured
MMLU-Pro87.1%#5 of 136HuggingFace MMLU-Pro leaderboard2026-10-13
Epoch ECI148#14 of 204Epoch AI Benchmarking Hub2026-10-13
Fiction.liveBench86.1%#4 of 47Epoch AI Benchmarking Hub2026-10-13
OSWorld63.3%#1 of 5Epoch AI Benchmarking Hub2026-10-13
GPQA Diamond87.6%#56 of 275HuggingFace GPQA Diamond leaderboard2026-10-13
Humanity's Last Exam24.4%#6 of 27Epoch AI Benchmarking Hub2026-10-13
SWE-bench Verified70.8%#17 of 73SWE-bench Verified leaderboard2026-10-13
CL-bench Life13.2%#3 of 9Epoch AI Benchmarking Hub2026-10-13
Terminal-Bench 2.043.2%#10 of 28Epoch AI Benchmarking Hub2026-10-13
WeirdML45.6%#40 of 109Epoch AI Benchmarking Hub2026-10-13
SciCode49.0%#39 of 106Epoch AI Benchmarking Hub2026-10-13
CL-bench19.3%#6 of 16Epoch AI Benchmarking Hub2026-10-13
SimpleBench46.8%#24 of 61Epoch AI Benchmarking Hub2026-10-13
EnigmaEval3.4%#10 of 24Epoch AI Benchmarking Hub2026-10-13
ALE-Bench822#33 of 79Epoch AI Benchmarking Hub2026-10-13
WebDev Arena1,436#31 of 67Epoch AI Benchmarking Hub2026-10-13
CritPt3.1%#55 of 115Epoch AI Benchmarking Hub2026-10-13
ARC-AGI-211.8%#48 of 87ARC Prize official leaderboard2026-10-13
SimpleQA Verified34.3%#42 of 69Epoch AI Benchmarking Hub2026-10-13
Vending-Bench 2$1,198#12 of 19Epoch AI Benchmarking Hub2026-10-13
SWE-bench Multilingual67.3%#7 of 11SWE-bench Multilingual leaderboard2026-10-13

About Kimi K2.5

Kimi K2.5 is an open-weight model that can run on your own hardware from Moonshot. It was released on 2026-01-27, according to Epoch AI. The cheapest API listing we track charges $0.450 per million input tokens and $2.25 per million output tokens, through deepinfra. We hold 21 benchmark results for it across 4 areas, and 3 are in the top tenth of their boards.

Where Kimi K2.5 stands

On our aggregate score, Kimi K2.5 is #173 of 743 ranked models, in the top quarter. Its score is 50.7; the scale is centred on 50, so it sits 0.7 points above the typical model. That beats 77% of the ranked field. Among the 13 ranked Moonshot models it is #3. The next one up is Kimi K2.5 (Fireworks) at 52.8. The next one down is Kimi K2 Thinking Turbo at 50.5.

Kimi K2.5, benchmark by benchmark

On MMLU-Pro, Kimi K2.5 scores 87.1% and ranks #5 of 136. Only 4 models score higher, led by Darwin-180B-RSI at 88.1%, which is 1.0 points ahead. It covers 57-subject knowledge + reasoning. The board median is 55.3%, so Kimi K2.5 is 31.8 points above it. Source: HuggingFace MMLU-Pro leaderboard, captured 2026-10-13.

On Epoch ECI, Kimi K2.5 scores 148 and ranks #14 of 204. Only 13 models score higher, led by Qwen 3.8 Max at 156, which is 8 points ahead. It covers Epoch Capabilities Index: Epoch's own aggregate of ~40 benchmarks. The board median is 127, so Kimi K2.5 is 21 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

On Fiction.liveBench, Kimi K2.5 scores 86.1% and ranks #4 of 47. Only 3 models score higher, led by grok-4-0709 at 94.4%, which is 8.3 points ahead. It covers long-context story comprehension (16k). The board median is 61.1%, so Kimi K2.5 is 25.0 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

GPQA Diamond puts Kimi K2.5 at 87.6%, #56 of 275. That beats 80% of the models tested, but 8.2 points short of GPT-6 Astra (max) (95.8%). The results either side are GPT-5.1 (high) (87.6%) and Kimi K2.5 (Fireworks) (87.6%). It covers expert-level science questions. The board median is 71.7%, so Kimi K2.5 is 15.9 points above it. Source: HuggingFace GPQA Diamond leaderboard, captured 2026-10-13.

On Humanity's Last Exam, Kimi K2.5 reaches 24.4%. It ranks #6 of 27, ahead of 78% of the field, and 16.2 points behind Muse Spark at 40.6%. The results either side are GPT-5 (high) (25.3%) and Gemini 2.5 Pro Preview (Jun 2025) (21.6%). It covers expert-level questions across domains. The board median is 10.7%, so Kimi K2.5 is 13.7 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

Kimi K2.5 manages 70.8% on SWE-bench Verified, good for #17 of 73. Claude Opus 4.7 (max) leads at 83.5%, 12.7 points further up, and 77% of models sit below. The results either side are Claude 4 Sonnet (70.8%) and Claude Opus 4 (70.7%). It covers real GitHub issues resolved. The board median is 56.4%, so Kimi K2.5 is 14.4 points above it. Source: SWE-bench Verified leaderboard, captured 2026-10-13.

Kimi K2.5 lands below the middle on ARC-AGI-2: 11.8%, #48 of 87, with 54% of the field scoring higher. GPT-6 Astra (max) leads at 95.0%; the closest result above is Opus 4.5 (Thinking, 8K) at 13.9%. It covers abstract reasoning, ARC Prize public eval set. The board median is 22.8%, so Kimi K2.5 is 11.0 points below it. Source: ARC Prize official leaderboard, captured 2026-10-13.

Kimi K2.5 lands below the middle on SimpleQA Verified: 34.3%, #42 of 69, with 59% of the field scoring higher. gemini-4-argon-high leads at 76.0%; the closest result above is Kimi K2.6 at 34.9%. It covers short factual questions. The board median is 41.0%, so Kimi K2.5 is 6.7 points below it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

Kimi K2.5 lands below the middle on Vending-Bench 2: $1,198, #12 of 19, with 58% of the field scoring higher. Kimi K2.6 leads at $6,205; the closest result above is MiniMax-M3 at $2,158. It covers running a vending business for a year (final money, $). The board median is $2,377, so Kimi K2.5 is 1,178 points below it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

Kimi K2.5 lands below the middle on SWE-bench Multilingual: 67.3%, #7 of 11, with 55% of the field scoring higher. Gemini 3 Flash leads at 72.7%; the closest result above is MiniMax-M2.5 at 68.3%. It covers real GitHub issues resolved. The board median is 68.3%, so Kimi K2.5 is 1.0 points below it. Source: SWE-bench Multilingual leaderboard, captured 2026-10-13.

The full table above also lists the other 11 results, including boards with fewer than ten models, where a rank says little.

Strengths and weaknesses

Software engineering: solid, ahead of 63% of the field on average across 5 benchmarks. Its best board here is SWE-bench Verified (#17 of 73) and its weakest is WebDev Arena (#31 of 67). Reasoning, maths and science: solid, ahead of 67% of the field on average across 7 benchmarks. Its best board here is MMLU-Pro (#5 of 136) and its weakest is ARC-AGI-2 (#48 of 87). General knowledge and chat: solid, ahead of 60% of the field on average across 7 benchmarks. Its best board here is Epoch ECI (#14 of 204) and its weakest is SWE-bench Multilingual (#7 of 11).

Price and value

The median input price across the 427 priced models is $0.500 per million tokens, so Kimi K2.5 costs less than the typical listing. Models that score within 1.5 points or better and cost less: Qwen2.5-Coder-7B-Instruct ($0.010, 49.5), Qwen3.7 Flash ($0.030, 50.5), GLM-5.2 (high) ($0.060, 51.6). Similar scores cost more on Gemini 3 Flash ($0.500) and SWE-1.7 ($0.500).

Release history

The Moonshot model released before it was Kimi-K2-Instruct-0905 on 2025-09-05, which scores 49.5. The next one was Kimi K2.5 (Fireworks) on 2026-01-27, at 52.8.

Reading these numbers

The 21 scores on this page come from 6 sources: Epoch AI Benchmarking Hub (16), ARC Prize official leaderboard (1), HuggingFace GPQA Diamond leaderboard (1), HuggingFace MMLU-Pro leaderboard (1). 2 of them come from an official leaderboard and 19 from datasets or third-party runs.

Models ranked nearby

More from Moonshot

Compare it on the live leaderboard → · How the aggregate score works