vs
0:8
overall capability wins (0 ties across 8 scored). Kimi K3 leads overall.
- Reasoning0:4
- Coding0:1
- Pricing3:0
- Speed1:1
- Specs0:3
Verdict
Kimi K3 leads coding (widest gap: +25.0 on SciCode); Kimi K3 offers ~8.2x larger context (1.0M vs 128K); Kimi K3 allows ~7.8x more max output (64K vs 8K).
- Kimi K3 leads coding (widest gap: +25.0 on SciCode)
- Kimi K3 offers ~8.2x larger context (1.0M vs 128K)
- Kimi K3 allows ~7.8x more max output (64K vs 8K)
- Kimi K2 is ~5.6x 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
- $1.23
- Kimi K3
- $6.75
Kimi K2 is about 5.5x 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
Jul 11, 2025
Jul 16, 2026
Knowledge cutoff
-
-
API / provider
Moonshot
Moonshot
Modalities
text → text
text, image, video → text
Specs
Context window
128K
1.0M
Max output
8K
64K
Parameters
1.0T (32B act.)
2.8T (104B act.)
Pricing
Input $/1M
$0.60
$3
Output $/1M
$2.50
$15
Blended $/1M (3∶1)
$1.07
$6
Speed
tok/s
50
60
TTFT (s)
0.55
0.6
Reasoning
MMLU-Pro
81.1
—
GPQA Diamond
76.6
93.5
Humanity's Last Exam
7.4
46.9
AIME 2025
49.5
—
MATH-500
97.4
—
Humanity's Last Exam (with tools)
—
—
AA-Omniscience Accuracy
27.4
47.6
AA-LCR v1.1
53.0
88.7
CritPt
—
23.4
MMMU-Pro
—
80.5
IFBench
41.5
—
Chartography
—
—
Chartography (With Tools)
—
—
Coding
SWE-bench Verified
65.8
—
SWE-bench Pro
27.7
—
SWE-bench Multilingual
47.3
—
LiveCodeBench
53.7
—
Terminal-Bench 2.1
—
85.0
Aider Polyglot
59.1
—
Terminal-Bench 3
—
—
BigCodeBench
—
—
SciCode
34.5
59.5
CursorBench
—
60.8
SWE-Rebench
34.6
—
NL2Repo-Bench
—
—
DeepSWE
—
69.0
WebDev Arena
—
1,674
Terminal-Bench 4.0
—
—
CursorBench 4.0
—
—
FrontierCode v1.1 (Main)
—
—
Arena
LMArena Elo
—
1,485
| Metric | Kimi K2 | Kimi K3 | Delta |
|---|---|---|---|
| Identity | |||
| Organization | Moonshot | Moonshot | — |
| Family | Kimi | Kimi | — |
| License | Modified MIT | Kimi K3 License | — |
| Open weights | Yes | Yes | — |
| Release | Jul 11, 2025 | Jul 16, 2026 | — |
| Knowledge cutoff | - | - | — |
| API / provider | Moonshot | Moonshot | — |
| Modalities | text → text | text, image, video → text | — |
| Specs | |||
| Context window | 128K | 1.0M | B +921K |
| Max output | 8K | 64K | B +56K |
| Parameters | 1.0T (32B act.) | 2.8T (104B act.) | B +2K |
| Pricing | |||
| Input $/1M | $0.60 | $3 | A +$2.40 |
| Output $/1M | $2.50 | $15 | A +$12.50 |
| Blended $/1M (3∶1) | $1.07 | $6 | A +$4.92 |
| Speed | |||
| tok/s | 50 | 60 | B +10 |
| TTFT (s) | 0.55 | 0.6 | A +0.05 |
| Reasoning | |||
| MMLU-Pro | 81.1 | — | — |
| GPQA Diamond | 76.6 | 93.5 | B +16.9 pts |
| Humanity's Last Exam | 7.4 | 46.9 | B +39.5 pts |
| AIME 2025 | 49.5 | — | — |
| MATH-500 | 97.4 | — | — |
| Humanity's Last Exam (with tools) | — | — | — |
| AA-Omniscience Accuracy | 27.4 | 47.6 | B +20.2 pts |
| AA-LCR v1.1 | 53.0 | 88.7 | B +35.7 pts |
| CritPt | — | 23.4 | — |
| MMMU-Pro | — | 80.5 | — |
| IFBench | 41.5 | — | — |
| Chartography | — | — | — |
| Chartography (With Tools) | — | — | — |
| Coding | |||
| SWE-bench Verified | 65.8 | — | — |
| SWE-bench Pro | 27.7 | — | — |
| SWE-bench Multilingual | 47.3 | — | — |
| LiveCodeBench | 53.7 | — | — |
| Terminal-Bench 2.1 | — | 85.0 | — |
| Aider Polyglot | 59.1 | — | — |
| Terminal-Bench 3 | — | — | — |
| BigCodeBench | — | — | — |
| SciCode | 34.5 | 59.5 | B +25.0 pts |
| CursorBench | — | 60.8 | — |
| SWE-Rebench | 34.6 | — | — |
| NL2Repo-Bench | — | — | — |
| DeepSWE | — | 69.0 | — |
| WebDev Arena | — | 1,674 | — |
| Terminal-Bench 4.0 | — | — | — |
| CursorBench 4.0 | — | — | — |
| FrontierCode v1.1 (Main) | — | — | — |
| Arena | |||
| LMArena Elo | — | 1,485 | — |
Kimi K2
- Context
- 128K / 8K out
- Parameters
- 1.0T (32B act.)
- Price
- $0.60 / $2.50
- Speed
- 50 tok/s · 0.55s TTFT
- Modalities
- text → text
- License
- Modified MIT
Moonshot's trillion-parameter MoE (32B active) — open agentic coding before K2.5 / K3.
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.