Kimi K2 Instruct 0905 — Benchmarks

Benchmark scores for Kimi K2 Instruct 0905 aggregated from public leaderboards (7 boards), with how it ranks among open models and how it compares with its nearest neighbours. See hardware requirements for what you need to run it.

llmrun Composite Score →

102.7

90% range 98.6–106.8 · from 6 boards

Coding100.090% range 95–105
AgentsNot enough data yet
MathNot enough data yet
ScienceNot enough data yet
ReasoningNot enough data yet

Kimi K2 Instruct 0905 reasoning benchmarks

Multi-step logic and problem-solving — can the model think through a hard problem rather than recall an answer.

Fiction.LiveBench
Fiction.LiveBench is a long-context comprehension test: models answer questions that require tracking characters, events and relationships across a long story. llmrun shows the 16k-token score, the column Epoch AI reports as its headline.
66.7%
66.7%
median 61.1%open leader 86.1%
Overall best: GPT 5 (Aug 07, 2025, medium) 97.2%
63rd percentile · #10 of 25 open · #26 of 58 all
#10 of 25
#26 of 58
Independent
Epoch AI · Sep 2025
ForecastBench
ForecastBench, from the Forecasting Research Institute, asks models to predict real-world future events, with questions that only resolve after the model's training cutoff so the answers can't have been memorised. It is scored as a difficulty-adjusted Brier index, where higher is better.
62.7
62.7
median 61.5open leader 64.7
Overall best: GPT 5.5 (high) 67.0
83rd percentile · #5 of 24 open · #57 of 92 all
#5 of 24
#57 of 92
Lab-reported
Epoch AI · Sep 2025

Kimi K2 Instruct 0905 coding benchmarks

Writing, editing, and fixing real code — measured by whether the resulting program actually runs and passes tests.

SWE-bench Verified
SWE-bench Verified tests whether a model can resolve real GitHub issues from popular open-source Python projects, scored on the official swebench.com leaderboard as the percentage of human-validated issues actually fixed. It is the headline measure of practical, agentic software-engineering ability — where open-weight models like Qwen3-Coder, GLM-4.6, Kimi K2 and DeepSWE are now competitive with the frontier.
71.2%
71.2%
median 62.2%open leader 75.8%
Overall best: live-SWE-agent + Claude 4.5 Opus medium (20251101) 79.2%
89th percentile · #3 of 20 open · #33 of 162 all
#3 of 20
#33 of 162
Independent
SWE-bench · Oct 2025
Aider Polyglot
The Aider Polyglot benchmark measures real-world coding across several programming languages: the model edits code to solve Exercism exercises, and is scored on whether the final solution actually runs and passes the tests.
59.1%
59.1%
median 45.1%open leader 74.2%
Overall best: GPT 5 (Aug 07, 2025, high) 88.0%
74th percentile · #6 of 18 open · #30 of 69 all
#6 of 18
#30 of 69
Independent
Epoch AI · Sep 2025
WeirdML
WeirdML asks models to write machine-learning code for small, deliberately unusual datasets, then runs the code and scores the accuracy it actually achieves. Built by Håvard Tveit Ihle, it rewards practical problem-solving over recognizing a familiar task.
36.7%
36.7%
median 39.5%open leader 82.6%
Overall best: GPT 6 Astra Promax 93.6%
33rd percentile · #32 of 47 open · #125 of 162 all
#32 of 47
#125 of 162
Independent
Epoch AI · Sep 2025
ALE-Bench
ALE-Bench, from Sakana AI and AtCoder, tests long-horizon algorithmic optimisation using problems drawn from AtCoder Heuristic Contests, where there is no single correct answer and solutions are scored on quality. Results are reported as an AtCoder-style performance rating, where higher is better.
267.1
267.1
median 754.6open leader 1932.8
Overall best: GPT 6 Astra Max 2951.3
6th percentile · #45 of 48 open · #122 of 129 all
#45 of 48
#122 of 129
Lab-reported
Epoch AI · Sep 2025

Kimi K2 Instruct 0905 instruction following benchmarks

How precisely a model obeys explicit, verifiable instructions — format, length, and content constraints — and, more broadly, how much people prefer its responses head-to-head in blind human voting.

LMArena Text
Arena (arena.ai, formerly LMArena / Chatbot Arena) ranks models by millions of blind, head-to-head human votes on real prompts. The Arena Score shown here is a Bradley-Terry rating computed with Arena's default style control, so higher means people preferred that model's answers more often — it's a relative rating, not a percentage, so compare it only to other models on this board.
1418.5
1418.5
median 1321.4open leader 1488.2
Overall best: Gemini 4 Carbon (High) 1527.5
82nd percentile · #35 of 188 open · #129 of 408 all
#35 of 188
#129 of 408
Arena
Arena · Oct 2026

Score vs VRAM in the ≤640 GB tier

Kimi K2 Instruct 0905
Other models
Best Score for its VRAM
x: VRAM (GB, log scale) · y: llmrun Score

#47 of 102 that fit ≤640 GB

Each dot is a model at the best quant that fits ≤640 GB. The line steps up to the best Score available at each amount of VRAM; a dot below it is beaten by a model that needs no more memory.

Full ≤640 GB ranking →

Kimi K2 Instruct 0905 vs its neighbours

Same boards for the nearest open models per llmrun Score. The highest score in each row is marked.

Board	Kimi K2 Instruct 0905this model	Qwen3.8 27Bbest in classQ4_K_M · 18 GB	DeepSeek R1 0528nearest by ScoreQ4_K_M · 412 GB	DeepSeek V3.1nearest by ScoreQ4_K_M · 412 GB	Kimi K2 Instructnearest by ScoreQ4_K_M · 617 GB
llmrun Score	102.7	122.0	103.7	100.2	100.1
Fiction.LiveBench	66.7%	—	75.0%	52.8%	61.1%
ForecastBench	62.7	—	—	60.7	62.2
SWE-bench Verified	71.2%	—	—	—	65.4%
Aider Polyglot	59.1%	—	71.4%	—	59.1%
WeirdML	36.7%	—	41.6%	38.4%	39.4%
ALE-Bench	267.1	—	804.1	—	—
LMArena Text	1418.5	1435.6	1421.5	1416.9	1417.8
Compare head to head		Compare	Compare	Compare	Compare

Scores aggregated from public benchmark sources (Epoch AI: 5 boards, Arena: 1 board, SWE-bench: 1 board) — each linked from the benchmark pages. llmrun does not run these benchmarks. Percentiles and ranks are among open models in the llmrun catalogue; the median and leader marks refer to the same field.