kimi-k2.6-thinking: benchmark scores, price and ranking

kimi-k2.6-thinking from Moonshot AI ranks #379 of 743 LLMs on the LLMs Tiger aggregate benchmark score (48.2), based on 7 benchmark results captured up to 2026-10-13.

Organisation

Moonshot AI

Type

Open-weight / local model

Open weights

Yes

Cheapest API price

$0.650 in / $3.41 out per million tokens (via openrouter, observed 2026-10-13T20:47Z)

Aggregate rank

#379 of 743 (score 48.2)

Relative to other models, kimi-k2.6-thinking is strongest on LiveBench Coding (#5 of 17) and weakest on LiveBench Data Analysis (#15 of 17).

Across the 7 benchmarks where at least ten models are scored, kimi-k2.6-thinking is in the top 10% on 0 and in the top half on 2.

Its cheapest listed input price, $0.650 per million tokens, is lower than that of 43% of the 427 models with a listed price.

kimi-k2.6-thinking benchmark results

About kimi-k2.6-thinking

kimi-k2.6-thinking is an open-weight model that can run on your own hardware from Moonshot AI. The cheapest API listing we track charges $0.650 per million input tokens and $3.41 per million output tokens, through openrouter. We hold 7 benchmark results for it across 3 areas, and none is in the top tenth of its board.

Where kimi-k2.6-thinking stands

On our aggregate score, kimi-k2.6-thinking is #379 of 743 ranked models, around the middle of the table. Its score is 48.2; the scale is centred on 50, so it sits 1.8 points below the typical model. That beats 49% of the ranked field. Among the 3 ranked Moonshot AI models it is #3. The next one up is kimi-k2p5 at 49.2.

kimi-k2.6-thinking, benchmark by benchmark

LiveBench Coding puts kimi-k2.6-thinking at 78.6%, #5 of 17. That beats 71% of the models tested, but 5.1 points short of GPT-5.2 Codex (83.6%). The results either side are claude-opus-4-8-xhigh-effort (79.3%) and gemini-3.5-flash-high (78.2%). It covers LiveBench release 2026_06_25. The board median is 76.5%, so kimi-k2.6-thinking is 2.1 points above it. Source: LiveBench 2026_06_25 official CSV, captured 2026-10-13.

On LiveBench Agentic Coding, kimi-k2.6-thinking reaches 46.9%. It ranks #7 of 17, ahead of 59% of the field, and 9.1 points behind claude-opus-4-8-xhigh-effort at 56.1%. The results either side are gemini-3.5-flash-high (49.0%) and GPT-5.4 nano (xhigh) (46.8%). It covers LiveBench release 2026_06_25. The board median is 45.8%, so kimi-k2.6-thinking is 1.1 points above it. Source: LiveBench 2026_06_25 official CSV, captured 2026-10-13.

kimi-k2.6-thinking lands below the middle on LiveBench IF: 64.4%, #10 of 17, with 53% of the field scoring higher. gemini-3.1-pro-preview-high leads at 79.1%; the closest result above is grok-build-0.1 at 65.2%. It covers LiveBench release 2026_06_25. The board median is 65.2%, so kimi-k2.6-thinking is 0.9 points below it. Source: LiveBench 2026_06_25 official CSV, captured 2026-10-13.

kimi-k2.6-thinking lands below the middle on LiveBench Language: 75.1%, #11 of 17, with 59% of the field scoring higher. GPT-5.5 (xhigh) leads at 87.4%; the closest result above is MiniMax-M3 at 76.8%. It covers LiveBench release 2026_06_25. The board median is 77.9%, so kimi-k2.6-thinking is 2.8 points below it. Source: LiveBench 2026_06_25 official CSV, captured 2026-10-13.

kimi-k2.6-thinking lands below the middle on LiveBench Mathematics: 84.3%, #11 of 17, with 59% of the field scoring higher. GPT-5.5 (xhigh) leads at 95.9%; the closest result above is Qwen3.7 Max at 85.2%. It covers LiveBench release 2026_06_25. The board median is 88.2%, so kimi-k2.6-thinking is 4.0 points below it. Source: LiveBench 2026_06_25 official CSV, captured 2026-10-13.

kimi-k2.6-thinking lands below the middle on LiveBench Reasoning: 79.4%, #11 of 17, with 59% of the field scoring higher. claude-opus-4-8-xhigh-effort leads at 89.7%; the closest result above is claude-opus-4-5-20251101-thinking-64k-high-effort at 80.1%. It covers LiveBench release 2026_06_25. The board median is 81.1%, so kimi-k2.6-thinking is 1.7 points below it. Source: LiveBench 2026_06_25 official CSV, captured 2026-10-13.

LiveBench Data Analysis is a weak spot. kimi-k2.6-thinking scores 65.1%, #15 of 17, behind 82% of the models measured. The leader, GPT-5.5 (xhigh), reaches 81.6%. It covers LiveBench release 2026_06_25. The board median is 71.8%, so kimi-k2.6-thinking is 6.7 points below it. Source: LiveBench 2026_06_25 official CSV, captured 2026-10-13.

Strengths and weaknesses

Software engineering: solid, ahead of 65% of the field on average across 2 benchmarks. Its best board here is LiveBench Coding (#5 of 17) and its weakest is LiveBench Agentic Coding (#7 of 17). Reasoning, maths and science: uneven, ahead of 35% of the field on average across 2 benchmarks. Its best board here is LiveBench Mathematics (#11 of 17) and its weakest is LiveBench Mathematics (#11 of 17). General knowledge and chat: weak, ahead of 29% of the field on average across 3 benchmarks. Its best board here is LiveBench IF (#10 of 17) and its weakest is LiveBench Data Analysis (#15 of 17).

Price and value

The median input price across the 427 priced models is $0.500 per million tokens, so kimi-k2.6-thinking costs more 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), DeepSeek V4.1 Flash (high) ($0.010, 48.9), DeepSeek V4 Flash 0731 (max) ($0.016, 53.1). Similar scores cost more on Kimi K2.7 Code ($0.671) and Mistral Large 4 (high) ($0.680).

Reading these numbers

The 7 scores on this page come from 1 source: LiveBench 2026_06_25 official CSV (7). 7 of them come from an official leaderboard and 0 from datasets or third-party runs. With 7 results the aggregate rank can move a lot when one new score arrives.

Models ranked nearby

• #376 qwen2:72b-instruct-q8_0 (48.2)
• #377 Qwen3-Max-Instruct (48.2)
• #378 internlm3-8b-instruct (48.2)
• #380 o1 (high) (48.1)
• #381 llama3-70b-8192 (48.1)
• #382 gemini-1.5-pro-001-feb24 (48.1)

More from Moonshot AI

• Kimi K2 (#98, 51.8)
• kimi-k2p5 (#289, 49.2)

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