Kimi K2 Thinking
Moonshot AIfamily · Kimihuggingface.co/moonshotai/Kimi-K2-Thinking
Kimi K2 Thinking is Moonshot AI’s most advanced open reasoning model to date, extending the K2 series into agentic, long-horizon reasoning. Built on the trillion-parameter Mixture-of-Experts (MoE) architecture introduced in...
Updated 4 h ago · first seen 11 Sept 2026
model_01M294WW95XPFV3AGY9BV6JMYW
Overview
Identity
- Canonical model
- Yesidentity confidence: mediumOne row per real model release. Artifacts (checkpoints, quantisations, conversions) and folded evaluation variants point here.
- Official checkpoints
- official_checkpoints = hf_repo identifiers carried by the model itself; artifacts are separate entities pointing here through canonical_id.
- Artifacts
- None recordedSeparate entities (checkpoint · quantization · conversion · packaging) pointing to this model through canonical_id.
- Provider deployments
- 2
- API aliases
- kimi-k2-thinkingmoonshotai/kimi-k2-thinkingIdentifiers under which providers and evaluators refer to this model.
- Folded evaluation variants
- 0Effort / thinking variants (…-high, …-non-reasoning) are result configurations of this model, not separate models. Their old URLs redirect here.
Openness
Open weights— weights downloadable under Other; 7 dimensions unknown.
Weights downloadable under a permissive or Creative Commons licence allowing commercial use; code or data may be missing.
Weights
Yes
Inference code
—
Training code
—
Training data
—
Dataset
—
Commercial use
—
Redistribution
—
Derivatives
—
Licence: Other (unclassified licence) (unknown · stated as “other”)
dimensions marked null are unknown, not false
Key facts
- Release date
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 5 d agomedium
- Status
Source:Artificial AnalysisT2observed 5 d agomedium
- Model card
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 5 d agomedium
- Openrouter id
Source:OpenRouter public model & pricing listingT2observed 5 d agomedium
Architecture
- Architecture
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 5 d agomedium
- Model type
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 5 d agomedium
- Parameters
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 5 d agomedium
- Weights dtype
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 5 d agomedium
- File size
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 5 d agomedium
- Library name
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 5 d agomedium
- Pipeline tag
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 5 d agomedium
- Hugging Face repo
Source:OpenRouter public model & pricing listingT2observed 5 d agomedium
Capabilities
Modalities
- Modalities
- text
- Input
- text
- Output
- text
Capabilities
Tool calling
Yes
OpenRouter public model & pricing listing · T2
Structured output
Yes
OpenRouter public model & pricing listing · T2
Reasoning
Yes
OpenRouter public model & pricing listing · T2
Vision
Unavailable
Audio
Unavailable
Fine-tuning available
Unavailable
- Context window
Source:OpenRouter public model & pricing listingT2observed 5 d agomedium
- Max output
Source:OpenRouter public model & pricing listingT2observed 3 d agomedium
Benchmarks13
Compare with another model →Comparable same task and conditions · Partially comparable same task, conditions differ (effort, temperature, judge) · Not comparable different variant or metric
Current rows only, grouped by benchmark → canonical metric → comparability group (task configuration). Effort variants folded into this model appear as rows of the same group. 13 current rows in total. “vs leader” compares with the current leader of the benchmark's primary group only; other groups are not directly comparable. Comparability rules →
Providers & Pricing2
All offers in the price terminal →USD per 1M tokens as published by each provider; native units (per-request fees, flex/priority tiers) are kept verbatim. Rows are append-only — every price change is kept in the history below. Cost of a workload →
Price history
Output price · USD / 1M tokens 2 providers
- Moonshot AI Platform
- OpenRouter
- OpenRouterfirst observed $2.512 Sept 2026
- Moonshot AI Platformfirst observed $2.511 Sept 2026
Input price · USD / 1M tokens 2 providers
- Moonshot AI Platform
- OpenRouter
- OpenRouterfirst observed $0.6012 Sept 2026
- Moonshot AI Platformfirst observed $0.6011 Sept 2026
Hardware fit37
Assumptions (7)
- Estimated, not measured: weights = parameters × bytes/param × 1.15 runtime overhead (or the observed artifact file size when one is recorded).
- bytes/param: 4bit = 0.5, 8bit = 1.0, fp16 = 2.0 (uniform quantization, no per-layer exceptions).
- KV cache: 2 × layers × kv_heads × head_dim × 2 bytes × context × batch when the architecture is known; otherwise 0.5 GB per 8 192 tokens (× batch), independent of architecture (GQA/MLA models need less).
- A model 'fits' when the estimate is at most the device memory minus 2 GB reserved for the OS and framework.
- Mixture-of-experts models are estimated on total parameters (all experts must be resident); active parameters are ignored.
- Device memory uses the largest configuration when several are listed (e.g. Apple silicon tiers).
- Multi-GPU: device memories are summed; interconnect bandwidth, tensor-parallel replication and pipeline bubbles are not modelled.
Versions & Artifacts0
Version history
Max output3 changes
11 Sept 2026→12 Sept 2026→12 Sept 2026→13 Sept 2026current
Context window2 changes
11 Sept 2026→11 Sept 2026→12 Sept 2026current
Licensefirst observation only
11 Sept 2026current
Opennessfirst observation only
11 Sept 2026current
Parametersfirst observation only
11 Sept 2026current
Statusfirst observation only
11 Sept 2026current
Each hop is a claim: click a value for its source, tier and observation time. Nothing is overwritten — a new observation closes the previous claim.
Artifacts 0
No artifact (checkpoint, quantisation, conversion or packaging) points to this model yet.
Timeline14
Full timeline →Kimi K2 Thinking: max output tokens changed from 100352 to 98304
Max output100.3K tokens→98.3K tokensopenrouterKimi K2 Thinking: max output tokens changed from 235929 to 100352
Max output235.9K tokens→100.3K tokensopenrouterOpenRouter changed pricing for Kimi K2 Thinking: $0.6 in / $2.5 out per 1M tokens → $0.6 in / $2.5 out per 1M tokens
$0.60 in / $2.5 out→$0.60 in / $2.5 outopenrouterKimi K2 Thinking: max output tokens changed from 100352 to 235929
Max output100.3K tokens→235.9K tokensopenrouterOpenRouter lists Kimi K2 Thinking at $0.6 in / $2.5 out per 1M tokens
openrouterKimi K2 Thinking scores 31.06% on Terminal-Bench
artificial_analysisKimi K2 Thinking scores 92.98% on τ²-bench
artificial_analysisKimi K2 Thinking scores 68.1% on IFBench
artificial_analysisKimi K2 Thinking scores 23.82% on Humanity's Last Exam
artificial_analysisKimi K2 Thinking scores 83.84% on GPQA Diamond
artificial_analysisKimi K2 Thinking scores 22.02 on Artificial Analysis Intelligence Index
artificial_analysisKimi K2 Thinking: context length changed from 256000 to 262144
Context window256K tokens→262.1K tokensopenrouterKimi K2 Thinking: release date changed from 2025-11-04 to 2025-11-06
Release date4 Nov 2025→6 Nov 2025openrouterKimi K2 Thinking: release date changed from 2025-11-06 to 2025-11-04
Release date6 Nov 2025→4 Nov 2025huggingface
Change history73
Viewing AI Atlas as of 16 Mar 2026 — attributes exactly as the atlas knew them on that day; later corrections are not shown.
Back to today →Kimi K2 Thinking was not yet in AI Atlas on 16 Mar 2026
Release daterelease_date6
Statusstatus1
Opennessopenness1
Licenselicense1
Architecturearchitecture1
Parametersparameter_count1
Context windowcontext_length3
Max outputmax_output_tokens4
Modalitiesmodalities1
Input modalitiesmodalities_input1
Output modalitiesmodalities_output1
Quantization formatquant_format1
File sizefile_size_gb1
Hugging Face repohf_repo1
Pipeline tagpipeline_tag1
Model cardmodel_card_url1
Downloadsmetric.downloads5
Likesmetric.likes3
Aa context windowaa_context_window1
Aa deprecatedaa_deprecated1
Aa opennessaa_openness1
Accessaccess1
Artifact kindartifact_kind1
Descriptiondescription1
Gatedgated1
Hf inference providershf_inference_providers1
Is quantizedis_quantized1
Last modifiedlast_modified1
Library namelibrary_name1
License keylicense_key1
License namelicense_name1
License rawlicense_raw1
Aa median output tokens per secondmetric.aa_median_output_tokens_per_second10
Downloads all timemetric.downloads_all_time5
Model typemodel_type1
Openrouter idopenrouter_id1
Openrouter listed atopenrouter_listed_at1
Reasoningreasoning1
Structured outputstructured_output1
Supported parameterssupported_parameters1
Tool callingtool_calling1
Weights availableweights_available1
Weights dtypeweights_dtype1
Claims are temporal and append-only: a new observation closes the previous claim (valid_to) instead of overwriting it. Conflicting claims from different sources are kept side by side and flagged — never averaged. Methodology →
Provenance
Attributed facts
45
Source tiers
T245
Freshest observation
5 h ago
Conflicts
None
Source documents 6
Tier 1 = official/primary, 2 = quality secondary, 3 = community, 4 = unverified. Every snapshot is archived; see all sources and the methodology.
Data quality (72/100) measures how well AI Atlas knows this entity — completeness, primary-source ratio, freshness, conflicts — never how good the model is. Methodology →