Patchwire

AI lab · 7 models tracked

Moonshot AI models

Every Moonshot AI model released in the last 18 months, with release dates, context windows, API prices and benchmark scores. Newest: Kimi K3, 3mo ago.

Compare Moonshot AI models → All labs
Models · 30 days
0 of 7
Latest release
3mo ago
Cheapest input / 1M
$0.24 Kimi K3
Largest context
1M Kimi K3

Release history

Timeline

Moonshot AI ships a new model about every 62 days. Each dot is a release; click one to open it.

Jul 2025Oct 2025Jan 2026Apr 2026Jul 2026 Kimi K2 0711 — Jul 11, 2025Kimi K2 0905 — Sep 4, 2025Kimi K2 Thinking — Nov 6, 2025Kimi K2.5 — Jan 27, 2026Kimi K2.6 — Apr 20, 2026Kimi K2.7 Code — Jun 12, 2026Kimi K3 — Jul 16, 2026

Lineup

All Moonshot AI models

ModelReleasedContextInput / 1MOutput / 1MGPQA Diamond
Kimi K3moonshotai/kimi-k3
ReasoningOpen weightsVision
 Jul 16, 2026 1M944K out $0.24 $11.94 93.1%
Kimi K2.7 Codemoonshotai/kimi-k2.7-code
ReasoningOpen weightsVision
 Jun 12, 2026 262K236K out $0.67 $3.35 87.9%
Kimi K2.6moonshotai/kimi-k2.6
ReasoningOpen weightsVision
 Apr 20, 2026 262K236K out $0.65 $3.41 90.8%
Kimi K2.5moonshotai/kimi-k2.5
ReasoningOpen weightsVision
 Jan 27, 2026 262K236K out $0.50 $2.50 —
Kimi K2 Thinkingmoonshotai/kimi-k2-thinking
ReasoningOpen weights
 Nov 6, 2025 262K98K out $0.60 $2.50 —
Kimi K2 0905moonshotai/kimi-k2-0905
Open weights
 Sep 4, 2025 262K98K out $0.60 $2.50 —
Kimi K2 0711moonshotai/kimi-k2
Open weights
 Jul 11, 2025 131K98K out $0.57 $2.30 —

Benchmarks

Moonshot AI's best scores

The top Moonshot AI model on each benchmark, and where it ranks among all 222 tracked models.

BenchmarkBest Moonshot AI modelScoreOverall rank
FrontierSWESoftware engineering Kimi K3 25.9% #12of 19
DeepSWESoftware engineering Kimi K3 68.5% #8of 26
FrontierCodeSoftware engineering Kimi K3 44.2% #15of 41
SWE-bench VerifiedSoftware engineering Kimi K2.6 76.6% #9of 26
Terminal-BenchSoftware engineering Kimi K2.5 43.2% #21of 40
GBAEvalSoftware engineering Kimi K3 48.4% #9of 23
SciCodeSoftware engineering Kimi K3 59.5% #9of 93
Aider PolyglotSoftware engineering Kimi K2 0905 59.1% #7of 11
WeirdML v3Software engineering Kimi K3 7.5% #14of 22
METR time horizonSoftware engineering Kimi K2 Thinking 54 min #11of 13
APEX-AgentsAgents & computer use Kimi K3 47.8% #28of 51
OSWorld 2.0Agents & computer use Kimi K2.6 4.6% #7of 9
OSWorldAgents & computer use Kimi K2.5 63.3% #3of 4
Vending-Bench 2Agents & computer use Kimi K2.6 $6,205 #19of 61
PostTrainBenchAgents & computer use Kimi K3 32.0% #5of 11
Fiction.LiveBenchWriting & content Kimi K2.5 86.1% #3of 15
GDP.pdfWriting & content Kimi K3 19.0% #24of 38
SimpleQA VerifiedWriting & content Kimi K3 50.6% #25of 66
WebDev ArenaDesign & visual Kimi K3 1674 #9of 88
Blueprint-Bench 2Design & visual Kimi K3 29.5% #19of 32
Furniture AssemblyDesign & visual Kimi K3 34.2% #20of 34
ExploitBenchCybersecurity Kimi K2.6 18.4% #5of 8
GPQA DiamondReasoning & knowledge Kimi K3 93.1% #17of 92
Humanity's Last ExamReasoning & knowledge Kimi K2.5 24.4% #15of 23
ARC-AGI-2Reasoning & knowledge Kimi K3 60.4% #33of 63
ARC-AGI-1Reasoning & knowledge Kimi K3 94.5% #16of 63
SimpleBenchReasoning & knowledge Kimi K3 60.7% #21of 43
CritPtReasoning & knowledge Kimi K3 23.4% #18of 96
Chess PuzzlesReasoning & knowledge Kimi K3 37.0% #38of 70
LMCAReasoning & knowledge Kimi K3 52.7% #20of 96
FrontierMathMaths Kimi K3 72.2% #23of 72
FrontierMath Tier 4Maths Kimi K3 39.0% #23of 62
ProofBenchMaths Kimi K3 87.0% #10of 70
OTIS Mock AIMEMaths Kimi K3 97.2% #29of 90

Directory

Other labs