VerdictPal
Join

Models / Moonshot

Kimi K2.7 Code

Kimi · Released 2026-06-12

Moonshot's June 2026 coding-specialist Kimi release. Kimi's docs position `kimi-k2.7-code` as its most capable coding model, with a 256K context window, multimodal tool examples, mandatory thinking mode, and stronger long-horizon coding reliability than K2.6.

Budget coding agent model for repo-scale refactors, code review, MCP tool workflows, and long debugging loops where K2.6 was close but too token-hungry.

#23 of 51 on AA Index · current snapshot
42 AA Index · 10 benchmark rows · 1 sourcesPosition is benchmark-specific — not a cross-family or cross-source ranking.
Composite53
Context262K
Input / 1M$0.95
Output / 1M$4.0
Knowledge cutoffopen
Output speed47 t/s
TTFT (default API)1.17s
Statussolid
ReasoningCodingAgenticLong context
Reasoning
90
Coding
48
Agentic
67
Long context
66

68average across 4 tested families

Editor's note
K2.7 Code is not a general-chat replacement for K2.6; it is a coding-agent branch. The strongest launch claims are still first-party, so the card stays solid rather than flagship until independent rows land in the ledger.
Benchmark placements

Where Kimi K2.7 Code places on each public benchmark source that publishes a row. Rank counts every model with a latest row in the same test — not a universal quality score.

BenchmarkFamilyRankScoreSourceDate
GPQA DiamondReasoning#26/ 6990Artificial Analysis2026-07-21
Terminal-BenchAgentic#26/ 6967Artificial Analysis2026-07-21
Artificial Analysis Intelligence IndexReasoning#31/ 6742Artificial Analysis2026-07-21
Humanity's Last ExamReasoning#31/ 6533Artificial Analysis2026-07-21
SciCodeCoding#33/ 6548Artificial Analysis2026-07-21
AA-LCRLong context#34/ 6466Artificial Analysis2026-07-21
AA time to first tokenPerformance#40/ 69Artificial Analysis2026-07-21
IFBenchReasoning#40/ 5763Artificial Analysis2026-07-21
τ³-Bench BankingAgentic#42/ 6418Artificial Analysis2026-07-21
AA output speedPerformance#56/ 69Artificial Analysis2026-07-21
Family context

The three highest-scoring carded models in each capability family. Where Kimi K2.7 Code shows up, it's highlighted.

ReasoningKimi K2.7 Code · 90
  1. 1
    Claude Mythos Preview
    95
  2. 2
    GPT-5.6 Sol
    94
  3. 3
    Kimi K3
    94
CodingKimi K2.7 Code · 48
  1. 1
    GPT 5.5
    96
  2. 2
    Claude Opus 4.8
    96
  3. 3
    Claude Opus 4.6
    96
AgenticKimi K2.7 Code · 67
  1. 1
    GPT 5.5
    100
  2. 2
    Claude Opus 4.8
    100
  3. 3
    Claude Fable 5
    100
Long contextKimi K2.7 Code · 66
  1. 1
    Kimi K3
    75
  2. 2
    GPT-5.1
    75
  3. 3
    GPT-5.6 Sol
    74
Newest receipts

Benchmark rows added to the public ledger for Kimi K2.7 Code in the last 120 days. Older rows live in the full table below.

10 recent
BenchmarkFamilyScoreSourceDays ago
AA-LCRLong context66.33%Artificial Analysis2
AA output speedPerformance47 t/sArtificial Analysis2
AA time to first tokenPerformance1.17sArtificial Analysis2
Artificial Analysis Intelligence IndexReasoning41.9%Artificial Analysis2
GPQA DiamondReasoning89.6%Artificial Analysis2
Humanity's Last ExamReasoning32.8%Artificial Analysis2
IFBenchReasoning63.13%Artificial Analysis2
SciCodeCoding47.5%Artificial Analysis2
τ³-Bench BankingAgentic18.14%Artificial Analysis2
Terminal-BenchAgentic67.42%Artificial Analysis2
Family coverage

A high composite that hides a weak family is a trap. These bars surface the families where this model hasn't been publicly tested, and where it leads.

Reasoning90
Coding48
Agentic67
Long context66
Price vs. performance
0255075100$0.10$0.30$1$3$10$30input USD / 1M tokens · log scale →← intelligence

Hover or tab any dot for name, score, and input price.

AnthropicOpenAIMistralAlibabaDeepSeekGoogleMoonshotMiniMaxxAIMetaZhipuXiaomiMeituanCohere

Full benchmark ledger

Every catalog benchmark for Kimi K2.7 Code. Scores link to the original source; gaps mean no public row exists yet.

10 sourced rows

10 of 32 catalog benchmarks have a sourced row for Kimi K2.7 Code.

Reasoning

Coding

Agentic

Long context

Performance

Changelog
  1. Released Kimi K2.7 Code with 256K context, mandatory thinking mode, multimodal tool examples, and reported 30% lower reasoning-token usage versus K2.6.
Sources