SI Index Data as of Oct 9, 2026

Kimi K2.5: Price, Context, Benchmarks, and Release Details

Moonshot AI provisional listing open weights · first seen Oct 8, 2026 · score computed Oct 9, 2026, 01:49 UTC
59.0
SI Score (method si-v3-retained-evidence-2)
#65 of 140 ranked
100% confidence 100 percent, Full confidence — 6 of 8 expected sources in
Coverage 83% of expected source weight · 100% confidence at 80% coverage

Pillar breakdown

Coding (weight 40 percent) 72.4
Math (weight 15 percent) —
Preference (weight 15 percent) 77.3
Reasoning (weight 30 percent) 37.3

Pillar weights: reasoning 30% · math 15% · coding 40% · preference 15%. Benchmark results use fixed 0–100 scales before averaging; incomplete evidence is shrunk toward 50.

Facts

Input price / 1M
$0.60LiteLLMFirst-party API Standard token rate; MIT LiteLLM transcription. Provider documentation: https://platform.moonshot.ai/docs/guide/kimi-k2-5-quickstart. Exact endpoint only; cache/batch/long-context rates excluded Retrieved Oct 9, 2026 · MIT
Open source ↗
Output price / 1M
$3.00LiteLLMFirst-party API Standard token rate; MIT LiteLLM transcription. Provider documentation: https://platform.moonshot.ai/docs/guide/kimi-k2-5-quickstart. Exact endpoint only; cache/batch/long-context rates excluded Retrieved Oct 9, 2026 · MIT
Open source ↗
Context window
262Kmodels.devPublished source fact Retrieved Oct 9, 2026 · MIT
Open source ↗
Max output
262Kmodels.devPublished source fact Retrieved Oct 9, 2026 · MIT
Open source ↗
Released
Jan 1, 2026models.devPublished source fact Retrieved Oct 9, 2026 · MIT
Open source ↗
Open weights
YesHugging Face HubPublic Hub repo with weight files; gating/repo upload date is not release date Retrieved Oct 9, 2026 · factual metadata; model-specific licenses
Open source ↗
License
otherHugging Face HubPublished source fact Retrieved Oct 9, 2026 · factual metadata; model-specific licenses
Open source ↗
Input modalities
text, image, videomodels.devPublished source fact Retrieved Oct 9, 2026 · MIT
Open source ↗

Benchmark results

9 results
Benchmark Raw result Normalized (0–100) Pillar
arc agi v1 public eval 73.1%ARC PrizeARC Prize steward-published result; exact edition/subset; publication time is HTTP Last-Modified of the aggregate export, not the evaluation date. [variant] kimi-k2.5Published Oct 6, 2026 Retrieved Oct 9, 2026 · factual citation
Open source ↗ 		73.1 reasoning
arc agi v1 semi private 65.3%ARC PrizeARC Prize steward-published result; exact edition/subset; publication time is HTTP Last-Modified of the aggregate export, not the evaluation date. [variant] kimi-k2.5Published Oct 6, 2026 Retrieved Oct 9, 2026 · factual citation
Open source ↗ 		65.3 reasoning
arc agi v2 public eval 12.1%ARC PrizeARC Prize steward-published result; exact edition/subset; publication time is HTTP Last-Modified of the aggregate export, not the evaluation date. [variant] kimi-k2.5Published Oct 6, 2026 Retrieved Oct 9, 2026 · factual citation
Open source ↗ 		12.1 reasoning
arc agi v2 semi private 11.8%ARC PrizeARC Prize steward-published result; exact edition/subset; publication time is HTTP Last-Modified of the aggregate export, not the evaluation date. [variant] kimi-k2.5Published Oct 6, 2026 Retrieved Oct 9, 2026 · factual citation
Open source ↗ 		11.8 reasoning
hle scale 24.4%Humanity’s Last ExamPotential contamination warning: This model was evaluated after the public release of HLE, allowing model builder access to the prompts and solutions. [variant] Published Feb 13, 2026 Retrieved Oct 9, 2026 · factual citation
Open source ↗ 		24.4 reasoning
lmarena text 1445.2 eloLMArena / ArenaPublished source fact [variant] text / overallPublished Oct 2, 2026 Retrieved Oct 9, 2026 · CC-BY-4.0
Open source ↗ 		77.3 preference
swe bench verified 73.8%Epoch AI BenchmarkingEpoch-owned evaluation mean score; scores only, no benchmark questions [variant] Published Feb 17, 2026 Retrieved Oct 9, 2026 · CC-BY
Open source ↗ 		73.8 coding
swe bench verified 70.8%Official model cards via models.devLab-reported; metric resolved; transcribed by MIT models.dev catalog; not independently evaluated [variant] Retrieved Oct 9, 2026 · factual citation; MIT transcription
Open source ↗ 		70.8 coding
swe bench verified 70.8%SWE-bench VerifiedSWE-bench published model plus agent result; harness retained, not a base model evaluation [variant] mini-SWE-agent; high; 2.0.0Published Feb 17, 2026 Retrieved Oct 9, 2026 · factual citation
Open source ↗ 		70.8 coding

Normalization uses fixed absolute 0–100 scales for each unit, independently of other models that have a result on each benchmark. Coverage and evidence breadth still affect the composite; compare the evaluation conditions before reading a small score gap as decisive. “lab-reported” marks the provider's own published figure.

Sources: in and pending

in Reported (6)

  • ARC Prize · arrived Oct 8, 2026
  • Epoch AI Benchmarking · arrived Oct 8, 2026
  • Humanity’s Last Exam · arrived Oct 8, 2026
  • Official model cards via models.dev · arrived Oct 8, 2026
  • LMArena / Arena · arrived Oct 8, 2026
  • SWE-bench Verified · arrived Oct 8, 2026

pending Awaiting (2)

  • LiveBench · carries 11% of expected weight
  • Terminal-Bench · carries 6% of expected weight

The confidence % rises as pending sources publish. Some sources never cover some models — that is why 100% confidence arrives at 80% of expected weight, not at full coverage.