vs
0:15
overall capability wins (1 ties across 16 scored). Kimi K2.5 leads overall.
- Reasoning0:7
- Coding0:6
- Speed0:2
- Specs0:2
Verdict
Kimi K2.5 leads coding (widest gap: +31.3 on LiveCode); Kimi K2.5 offers ~2.0x larger context (256K vs 128K); Kimi K2.5 allows ~7.8x more max output (64K vs 8K).
- Kimi K2.5 leads coding (widest gap: +31.3 on LiveCode)
- Kimi K2.5 offers ~2.0x larger context (256K vs 128K)
- Kimi K2.5 allows ~7.8x more max output (64K vs 8K)
Cost for 1M in + 250K out
Illustrative chat workload at primary-provider list prices.
- Kimi K2
- $1.23
- Kimi K2.5
- $1.23
Price per 1M tokens
Full comparison
Organization
Moonshot
Moonshot
Family
Kimi
Kimi
License
Modified MIT
Modified MIT
Open weights
Yes
Yes
Release
Jul 11, 2025
Jan 27, 2026
Knowledge cutoff
-
-
API / provider
Moonshot
Moonshot
Modalities
text → text
text, image → text
Specs
Context window
128K
256K
Max output
8K
64K
Parameters
1.0T (32B act.)
1.0T (32B act.)
Pricing
Input $/1M
$0.60
$0.60
Output $/1M
$2.50
$2.50
Blended $/1M (3∶1)
$1.07
$1.07
Speed
tok/s
50
70
TTFT (s)
0.55
0.45
Reasoning
MMLU-Pro
81.1
87.1
GPQA Diamond
76.6
87.9
Humanity's Last Exam
7.4
30.7
AIME 2025
49.5
96.1
MATH-500
97.4
—
Humanity's Last Exam (with tools)
—
—
AA-Omniscience Accuracy
27.4
35.2
AA-LCR v1.1
53.0
78.0
CritPt
—
3.1
MMMU-Pro
—
75.4
IFBench
41.5
70.2
Chartography
—
—
Chartography (With Tools)
—
—
Coding
SWE-bench Verified
65.8
76.8
SWE-bench Pro
27.7
50.7
SWE-bench Multilingual
47.3
67.3
LiveCodeBench
53.7
85.0
Terminal-Bench 2.1
—
45.7
Aider Polyglot
59.1
—
Terminal-Bench 3
—
—
BigCodeBench
—
—
SciCode
34.5
49.0
CursorBench
—
—
SWE-Rebench
34.6
49.0
NL2Repo-Bench
—
—
DeepSWE
—
—
WebDev Arena
—
—
Terminal-Bench 4.0
—
—
CursorBench 4.0
—
—
FrontierCode v1.1 (Main)
—
—
Arena
LMArena Elo
—
—
| Metric | Kimi K2 | Kimi K2.5 | Delta |
|---|---|---|---|
| Identity | |||
| Organization | Moonshot | Moonshot | — |
| Family | Kimi | Kimi | — |
| License | Modified MIT | Modified MIT | — |
| Open weights | Yes | Yes | — |
| Release | Jul 11, 2025 | Jan 27, 2026 | — |
| Knowledge cutoff | - | - | — |
| API / provider | Moonshot | Moonshot | — |
| Modalities | text → text | text, image → text | — |
| Specs | |||
| Context window | 128K | 256K | B +128K |
| Max output | 8K | 64K | B +56K |
| Parameters | 1.0T (32B act.) | 1.0T (32B act.) | tie |
| Pricing | |||
| Input $/1M | $0.60 | $0.60 | tie |
| Output $/1M | $2.50 | $2.50 | tie |
| Blended $/1M (3∶1) | $1.07 | $1.07 | tie |
| Speed | |||
| tok/s | 50 | 70 | B +20 |
| TTFT (s) | 0.55 | 0.45 | B +0.10 |
| Reasoning | |||
| MMLU-Pro | 81.1 | 87.1 | B +6.0 pts |
| GPQA Diamond | 76.6 | 87.9 | B +11.3 pts |
| Humanity's Last Exam | 7.4 | 30.7 | B +23.3 pts |
| AIME 2025 | 49.5 | 96.1 | B +46.6 pts |
| MATH-500 | 97.4 | — | — |
| Humanity's Last Exam (with tools) | — | — | — |
| AA-Omniscience Accuracy | 27.4 | 35.2 | B +7.8 pts |
| AA-LCR v1.1 | 53.0 | 78.0 | B +25.0 pts |
| CritPt | — | 3.1 | — |
| MMMU-Pro | — | 75.4 | — |
| IFBench | 41.5 | 70.2 | B +28.7 pts |
| Chartography | — | — | — |
| Chartography (With Tools) | — | — | — |
| Coding | |||
| SWE-bench Verified | 65.8 | 76.8 | B +11.0 pts |
| SWE-bench Pro | 27.7 | 50.7 | B +23.0 pts |
| SWE-bench Multilingual | 47.3 | 67.3 | B +20.0 pts |
| LiveCodeBench | 53.7 | 85.0 | B +31.3 pts |
| Terminal-Bench 2.1 | — | 45.7 | — |
| Aider Polyglot | 59.1 | — | — |
| Terminal-Bench 3 | — | — | — |
| BigCodeBench | — | — | — |
| SciCode | 34.5 | 49.0 | B +14.5 pts |
| CursorBench | — | — | — |
| SWE-Rebench | 34.6 | 49.0 | B +14.4 pts |
| NL2Repo-Bench | — | — | — |
| DeepSWE | — | — | — |
| WebDev Arena | — | — | — |
| Terminal-Bench 4.0 | — | — | — |
| CursorBench 4.0 | — | — | — |
| FrontierCode v1.1 (Main) | — | — | — |
| Arena | |||
| LMArena Elo | — | — | — |
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 K2.5
- Context
- 256K / 64K out
- Parameters
- 1.0T (32B act.)
- Price
- $0.60 / $2.50
- Speed
- 70 tok/s · 0.45s TTFT
- Modalities
- text, image → text
- License
- Modified MIT
Prior Kimi open MoE generation - strong long-context agentic coding before K3.
Benchmark charts
Winner bars are emphasized. Per-benchmark deltas sit above each chart.