Moonshot AIChina· released Apr 2026· open weights

Kimi K2.6

Name: Kimi K2.6
Author: Moonshot AI

Provisional: 17 comparable tests so far. Ranked once it reaches 6 across 3 counting categories.

Moonshot AI, China. One of 5 models from this lab on the Index.

Third-party publishedLow confidence (thin coverage)

Snapshot September 8, 2026

OpenCharts Index

58.5

Provisional: not enough comparable tests or counting categories to rank yet · 17 comparable tests · 2 counting categories

Drop any single test and the Index lands between 50.7 and 64.1.

Needs more results (3 of 8 so far)

Averaging 59.7 on the 3 it has; the score opens at 4 of 8.

Distance to today's frontier on the hardest tests. Not a measure of intelligence.

What it's known for

Kimi K2.6, in 17 results.

Strongest in coding (67.0) and math (49.9). It trails the frontier most in math, 50.1% behind on average. No comparable result yet in long context. Between 40% and 50% behind the best published results, on average.

Strongest category

Coding

Writing, fixing and shipping real code, judged by tests or users.

67.0

SWE-bench Verified76.7% · 92% of best
SciCode53.5% · 86% of best
CursorBench47.6% · 65% of best
WeirdML55.9% · 60% of best
Code Arena (WebDev)1,509 · 32% win rate vs leader

Mean of 5 of the category's 7 comparable tests. Counts toward the Index.

Best single result

Mock AIME 2024–2025

Solve competition mathematics problems that have a single exact integer answer.

96.1%

96.1% of the best (100.0%) · #17 of 78.

Olympiad-style problems need long, careful chains of reasoning, and there is no partial credit.

Where it trails

OSWorld 2.0

Complete a multi-hour workflow across real desktop and web applications, end to end, with the environment changing as you work.

4.6%

85.4% behind the best published result (31.4%) · #7 of 9.

Computer use is how an agent reaches every tool that has no API, and long workflows are where agents lose the thread.

Not measured yet

No comparable result yet in this category. A blank is a blank, never a zero, and it does not lower the Index.

Long context

Covered, but not averaged into the Index:

Reasoning2 of 6 tests · needs 3 · mean 58.9
Knowledgesingle-test category, shown beside the Index · 46.2
Agentic3 of 8 tests · needs 4 · mean 45.2
Multimodalsingle-test category, shown beside the Index · 85.6
Human preferencesingle-test category, shown beside the Index · 86.7

Every score, with its source

See it at work

What Kimi K2.6 was asked to do, and how it did.

Every test it has taken, by category, with the kind of task it faced, who produced the number and where the result landed against the best published one. The examples are original and illustrative, never items from the datasets themselves.

Human preference · mean 86.7 across 1 comparable test · a single-test category, shown beside the Index and never averaged into it

Text ArenaLeaderboard

Answer the same prompt as a rival model; real users vote blind on which answer they prefer.

1,461

43% expected win rate against the board leader (1,507)
#37 of 92

board 2026-09-02

Source: Arena

Every result is read against the best published one on its test: percent scores as a share of the best above chance, Arena ratings as the expected win rate against the board leader, open-ended values on a log scale. Sitting a harder exam never lowers a model. Every number links to the publisher that produced it.

Head to head

Kimi K2.6 against whoever you pick.

Choose a rival. Every counted test both have taken appears side by side as a share of the best published result on that test, with the category means above. A win is a higher share; a gap under 2 points is a tie, because one item on a 45-problem exam is 2.2 points and publishers' own standard errors are of that order.

Kimi K2.6

provisional·Index 58.5C-

0 : 11

wins · 0 ties · 11 shared

GPT-6 Astra

#1·Index 99.0A+

GPT-6 Astra comes out ahead on 11 of the 11 tests both have taken.

Category means

Reasoning
58.9
99.6
Knowledge
46.2
100.0
Coding
67.0
97.7
Math
49.9
99.8
Agentic
45.2
98.5
Multimodal
85.6
—
Human preference
86.7
—

Shared tests · biggest gaps first

ProofBenchMath
16.0%
99.0%
−83.0
FrontierMath Tier 4Math
25.6%
97.6%
−73.7
CritPtReasoning
8.0%
31.7%
−73.4
Code Arena (WebDev)Coding
1,509
1,797
−68.0
APEX-AgentsAgentic
18.9%
46.7%
−58.6
SimpleQA VerifiedKnowledge
34.9%
75.6%
−53.8
WeirdMLCoding
55.9%
92.9%
−39.9
FrontierMath (Tiers 1–3)Math
57.2%
93.7%
−39.0
GPQA DiamondReasoning
90.8%
95.8%
−7.0
SciCodeCoding
53.5%
56.5%
−4.8
Mock AIME 2024–2025Math
96.1%
100.0%
−3.9

Bars are distances to the best published result on each test (percent scores above chance, Arena ratings as win rate against the leader, open-ended values on a log scale). Raw values as the publishers report them.

Open GPT-6 Astra

Scores

Every test, every source

The best published run per benchmark, the raw value as the publisher reports it, who produced the number, how close it comes to the best published result on that test, and where the model sits among every model scored on it. Percent scores are read above chance, Arena ratings as a win rate against the board leader, open-ended values on a log scale. A test with too few models to compare against is shown but not counted.

Category means

Reasoning58.9 · 2/6
Knowledgebeside46.2 · 1/1
Coding67.0 · 5/7
Math49.9 · 4/5
Agentic45.2 · 3/8
Multimodalbeside85.6 · 1/1
Human preferencebeside86.7 · 1/1
Long context—

Mean · tests taken of the category's comparable tests. A faded category either rests on fewer than half of them or is a single-test category (marked “beside”), which is shown next to the Index and never averaged into it.

GPQA DiamondEpoch-runReasoning

90.8%

93.0% of the best (95.8%), above chance
69th percentile of 78 · n = 198

Source: Epoch AI — AI Benchmarking Hub

CritPtLeaderboardReasoning

8.0%

24.8% of the best (32.3%)
48th percentile of 78

Source: Epoch AI — AI Benchmarking Hub

SimpleQA VerifiedEpoch-runKnowledge

34.9%

46.2% of the best (75.6%)
34th percentile of 57 · n = 1,000

Source: Epoch AI — AI Benchmarking Hub