Kimi K2
Context 128K tokens
Top 1 cheapest providers (Adjusted $/task)
Within the active global provider, residency and confidentiality filters.
| # | Provider | Platform | Raw input $/1M | Raw output $/1M | Adjusted $/task |
|---|---|---|---|---|---|
| 1 | Novita | OpenRouter | $0.570 | $2.30 |
Composite
1 of 7 inputs6 radar axes: DesignArena's two boards share one20.8
dominance-adjusted from 28.9: a better-measured model that is at least as good on each of these inputs ranks above it
AA Coding —Coding Agent v1.4 —AA Intelligence 12.7AA Agentic —Epoch ECI —Software ECI —DesignArena —/—
Radar: percentile among all models measured on each input; a gap means not measured.
Benchmark sheet
22 of 140 registered benchmark versions · bars show the percentile among all models measured on each benchmark.
Agentic
Terminal-Bench Hard (AA)5315.9%
Terminal-Bench Hard (AA) pinned revision 74221fb ↗ · Published board — Tests a pinned 44-task hard subset of terminal-based work using Terminus 2.
Elimination Game Benchmark423
Elimination Game Benchmark published 2026-09-10 ↗ · Published board — Multi-player tournament where 8 LLM players converse, form alliances, and vote to eliminate each other, with final rankings scored via TrueSkill.
Coding
Aider Polyglot6759.1%
Aider Polyglot published 2026-09-10 ↗ · edit format: diff — Evaluates LLMs on 225 Exercism coding exercises across six languages, testing code editing without human intervention.
GSO software optimization, Opt@1 (UC Berkeley) vopt1-102334.9%
GSO software optimization, Opt@1 (UC Berkeley) vopt1-102 ↗ · OpenHands — An agent gets a real codebase and a performance test and must make the code as fast as an expert developer's optimization; 102 tasks across 10 codebases and 5 languages.
Instruction-following
IFBench (AA single-turn)4241.5%
IFBench (AA single-turn) published 2026-09-10 ↗ · Published board — Tests precise single-turn instructions with deterministic rule checks.
LisanBench word chains, v0.2.0 (Lisan al Gaib)52348.7
LisanBench word chains, v0.2.0 (Lisan al Gaib) v0.2.0 ↗ · Published board — A model builds the longest chain of English words it can, each differing from the last by one letter, with no repeats and only dictionary words, from 50 starting words; any broken rule ends the chain.
Knowledge
Humanity's Last Exam (AA text-only)477.4%
Humanity's Last Exam (AA text-only) published 2026-09-10 ↗ · Published board — Tests expert-level knowledge on AA's text-only Humanity's Last Exam subset.
AA-Omniscience Index57-28.3
AA-Omniscience Index published 2026-09-10 ↗ · Published board — Tests factual reliability while rewarding correct answers and penalizing hallucinations.
UGI NatInt9548.9
UGI NatInt published 2026-09-10 ↗ · Published board — A private-question evaluation of general knowledge and reasoning across textbook, popular-culture and world-model topics.
Long-context
AA-LCR v1.14953.0%
AA-LCR v1.1 v1.1 ↗ · Published board — Tests reasoning across multiple long documents with corrected answer keys and grading.
Math
AIME 2025 (AA) v20255457.0%
AIME 2025 (AA) v2025 ↗ · Published board — Advanced mathematical problem solving on AIME I and II 2025.
Reasoning
SimpleBench3826.3%
SimpleBench published 2026-09-10 ↗ · Published board — A multiple-choice text benchmark of over 200 questions covering spatio-temporal reasoning, social intelligence, and linguistic adversarial robustness, on which a non-specialized human baseline outperforms every tested LLM.
AA Intelligence Index5412.7
AA Intelligence Index published 2026-09-19 ↗ · Published board — Artificial Analysis publishes this board without a version number; we keep the date each result was retained.
Safety/Alignment
BullshitBench V1 (clear pushback)3820.0%
BullshitBench V1 (clear pushback) published 2026-09-10 ↗ · Published board — Whether a model rejects the broken premise of 55 deliberately nonsensical prompts (V1 question set) instead of answering them confidently.
BullshitBench V2 (clear pushback)1010.0%
BullshitBench V2 (clear pushback) published 2026-09-10 ↗ · Published board — Whether a model rejects the broken premise of 100 deliberately nonsensical prompts (V2 question set) instead of answering them confidently.
Science
CritPt (AA)240.0%
CritPt (AA) published 2026-09-10 ↗ · Published board — Tests research-level physics reasoning with Python, symbolic and numerical answers.
GPQA Diamond (AA)6076.6%
GPQA Diamond (AA) published 2026-09-10 ↗ · Published board — Tests graduate-level biology, physics and chemistry knowledge on the Diamond subset.
Tool-use
τ²-Bench Telecom (AA)5761.1%
τ²-Bench Telecom (AA) published 2026-09-10 ↗ · Published board — Tests dual-control telecom agents that coordinate tool use with a simulated user.
Uncensored
UGI Willingness513.20
UGI Willingness published 2026-09-10 ↗ · Published board — A private-question evaluation of how far a model follows challenging instructions before refusing or deviating.
UGI9656.5
UGI published 2026-09-10 ↗ · Published board — A private-question suite combining sensitive-topic knowledge with willingness to follow controversial instructions.
Writing
EQ-Bench Creative Writing v3961,663 Elo
EQ-Bench Creative Writing v3 v3 ↗ · Published board — A LLM-judged creative writing benchmark with vocabulary and GPT-slop controls.
EQ-Bench Longform Creative Writing v1.118868.1
EQ-Bench Longform Creative Writing v1.11 ↗ · Published board — LLM-judged benchmark in which models plan and write a short story/novella over 8x 1000-word turns, scored 0-100 across 14 rubric dimensions.
UGI Writing9652.5
UGI Writing published 2026-09-10 ↗ · Published board — A maintainer-owned writing evaluation considering intelligence, style, repetition and length adherence, informed by human preferences.
Unusual results
3 threshold-crossing signals flagged · 19 eligible benchmark families
How flags are calculated
Heuristic screen, not statistical significance: benchmark families are correlated and source uncertainty is unknown. Peer evidence needs ≥ 20 independently measured matched configurations from ≥ 10 distinct model families. A flag needs a directed population z-score of magnitude ≥ 1.5 and a gap of ≥ 1.5 from the leave-one-benchmark-family-out mean z in the same direction, over ≥ 5 other benchmark families.
- unusually strong UGI NatInt published 2026-09-10
Why
Observed: 48.87 pointsPeer mean 26.4323 · peer sd 11.33206 · n 74 · families 73Directed z 1.98 · baseline z 0.196 · gap 1.784 · profile n 1848.87 pointsmeasuredobserved 2026-09-10huggingface.co ↗Evidence
Axis: UGI NatInt · published 2026-09-10 · Published boardExact value:48.87pointsObserved: 2026-09-10T22:04:13Z · publication date: not recordedObservation id:public:46c72ee62edcd2f19e8d8f74 - unusually strong UGI published 2026-09-10
Why
Observed: 56.55 pointsPeer mean 30.75703 · peer sd 13.08304 · n 74 · families 73Directed z 1.971 · baseline z 0.196 · gap 1.775 · profile n 1856.55 pointsmeasuredobserved 2026-09-10huggingface.co ↗Evidence
Axis: UGI · published 2026-09-10 · Published boardExact value:56.55pointsObserved: 2026-09-10T22:04:13Z · publication date: not recordedObservation id:public:2a61954b274de5212753d847 - unusually strong UGI Writing published 2026-09-10
Why
Observed: 52.5 pointsPeer mean 31.68366 · peer sd 10.68331 · n 71 · families 70Directed z 1.948 · baseline z 0.197 · gap 1.751 · profile n 1852.5 pointsmeasuredobserved 2026-09-10huggingface.co ↗Evidence
Axis: UGI Writing · published 2026-09-10 · Published boardExact value:52.5pointsObserved: 2026-09-10T22:04:13Z · publication date: not recordedObservation id:public:22d040d7b0ba9e9f7586440b
Missing coverage · 122 benchmark versions
No result does not mean a zero, or that the model was never tested. Collection failures and disputed versions retain their distinct status.
Token offers by platform · 1 offers (Adjusted $/task)
Click any underlined price to see how it is estimated and where each input comes from. How we calculate adjusted cost.
1 offers within the active global filters; “—” means the catalog is active but no public token price is available.
OpenRouter (1)
| Novita | globalnovita/fp8 | $0.570 raw in $/1M | $2.30 raw out $/1M |