LLMcompare

Search

Search for a command to run...

A · open · Kimi

Kimi K2.7 Code

Moonshot

B · open · Kimi

Kimi K3

Moonshot

0:14

overall capability wins (0 ties across 14 scored). Kimi K3 leads overall.

  • Reasoning0:5
  • Coding0:4
  • Tool use0:2
  • Composite indices0:2
  • Pricing3:0
  • Speed1:1
  • Specs0:3

Verdict

Kimi K3 leads coding (widest gap: +201.6 on WebDev Elo); Kimi K3 offers ~4.0x larger context (1.0M vs 262K); Kimi K3 allows ~2.0x more max output (64K vs 33K).

  • Kimi K3 leads coding (widest gap: +201.6 on WebDev Elo)
  • Kimi K3 offers ~4.0x larger context (1.0M vs 262K)
  • Kimi K3 allows ~2.0x more max output (64K vs 33K)
  • Kimi K2.7 Code is ~3.5x cheaper on a blended token basis than Kimi K3

Cost for 1M in + 250K out

Illustrative chat workload at primary-provider list prices.

Kimi K2.7 Code
$1.95
Kimi K3
$6.75

Kimi K2.7 Code is about 3.5x cheaper on this workload.

Price per 1M tokens

Full comparison

Organization

Moonshot

Moonshot

Family

Kimi

Kimi

License

Modified MIT

Kimi K3 License

Open weights

Yes

Yes

Release

Jun 12, 2026

Jul 16, 2026

Knowledge cutoff

-

-

API / provider

Moonshot

Moonshot

Modalities

text, image, video → text

text, image, video → text

Specs

Context window

262K

1.0M

B +786K

Max output

33K

64K

B +31K

Parameters

1.0T (32B act.)

2.8T (104B act.)

B +2K

Pricing

Input $/1M

$0.95

$3

A +$2.05

Output $/1M

$4

$15

A +$11

Blended $/1M (3∶1)

$1.71

$6

A +$4.29

Speed

tok/s

55

60

B +5.00

TTFT (s)

0.5

0.6

A +0.10

Reasoning

MMLU-Pro

—

—

—

GPQA Diamond

89.6

93.5

B +3.9 pts

Humanity's Last Exam

35.0

46.9

B +11.9 pts

AIME 2025

—

—

—

MATH-500

—

—

—

Humanity's Last Exam (with tools)

—

—

—

AA-Omniscience Accuracy

39.6

47.6

B +8.0 pts

AA-LCR v1.1

79.3

88.7

B +9.4 pts

CritPt

10.0

23.4

B +13.4 pts

MMMU-Pro

—

80.5

—

IFBench

63.1

—

—

Chartography

—

—

—

Chartography (With Tools)

—

—

—

Coding

SWE-bench Verified

—

—

—

SWE-bench Pro

—

—

—

SWE-bench Multilingual

—

—

—

LiveCodeBench

—

—

—

Terminal-Bench 2.1

67.4

85.0

B +17.6 pts

Aider Polyglot

—

—

—

Terminal-Bench 3

—

—

—

BigCodeBench

—

—

—

SciCode

47.8

59.5

B +11.7 pts

CursorBench

49.7

60.8

B +11.1 pts

SWE-Rebench

—

—

—

NL2Repo-Bench

—

—

—

DeepSWE

—

69.0

—

WebDev Arena

1,473

1,674

B +202 Elo

Terminal-Bench 4.0

—

—

—

CursorBench 4.0

—

—

—

FrontierCode v1.1 (Main)

—

—

—

Arena

LMArena Elo

—

1,485

—

Kimi K2.7 Code

Context
262K / 33K out
Parameters
1.0T (32B act.)
Price
$0.95 / $4
Speed
55 tok/s · 0.5s TTFT
Modalities
text, image, video → text
License
Modified MIT

Open-weight coding specialist (1T MoE / 32B active) with text/image/video input — long-horizon agentic software engineering at $0.95/$4; HighSpeed SKU is $1.90/$8 at ~180 tok/s.

Kimi K3

Context
1.0M / 64K out
Parameters
2.8T (104B act.)
Price
$3 / $15
Speed
60 tok/s · 0.6s TTFT
Modalities
text, image, video → text
License
Kimi K3 License

Largest open-weight model (2.8T MoE / 104B active) — official card reports strong reasoning and agentic coding results.

Benchmark charts

Winner bars are emphasized. Per-benchmark deltas sit above each chart.

Reasoning & knowledge

GPQA: B +3.9HLE: B +11.9CritPt: B +13.4AA-LCR: B +9.4Omniscience Acc.: B +8.0

Coding

TermBench: B +17.6SciCode: B +11.7CursorBench: B +11.1WebDev Elo: B +202

Tool use & function calling

Toolathlon V.: B +18.5τ³-Banking: B +25.8

Composite indices

AA Index: B +23.9AA Index v4.3: B +17.5

Pick a different pair · Back to catalog