Kimi K3
moonshotai/kimi-k3
long-horizon coding & knowledge work with 1M context.
cmd --model moonshotai/kimi-k3
Intelligence index
59.7
Output speed
39.4 tok/s
Input
$3 /M
Output
$15 /M
Cache read
$0.30 /M
Agent-loop cost
$1.11 /M in
Context window
1M tokens
Released
July 16, 2026
Modalities
→
vs. the lineup
Kimi K3 beside its stablemates and nearest rivals. The ◆ marks the best value in each column across every row shown.
pin a rival:
| Model | Intelligence | Coding | Speed | Input $/M | Output $/M | Blended $/M | Context |
|---|---|---|---|---|---|---|---|
| Kimi K3 ◆ | 59.7◆ | 76.2◆ | 39.4◆ | $3◆ | $15◆ | $6◆ | 1M◆ |
| Muse Spark 1.2 | 56.8◆ | 72.2◆ | —◆ | $1.25◆ | $4.25◆ | $2◆ | 1.05M◆ |
| Kimi K2.6 | 45.1◆ | 61.8◆ | —◆ | $0.95◆ | $4◆ | $1.71◆ | 256K◆ |
| Kimi K2.7 Code | 43◆ | 60.8◆ | 40.5◆ | $0.95◆ | $4◆ | $1.71◆ | 256K◆ |
| Kimi K2.5 | 36◆ | 46.8◆ | —◆ | $0.60◆ | $3◆ | $1.20◆ | 256K◆ |
| Kimi K2.7 Code HighSpeed | —◆ | —◆ | —◆ | $1.90◆ | $8◆ | $3.42◆ | 262K◆ |
| DeepSeek V4 Pro (latest)pinned | 45.3◆ | 59.4◆ | 63.3◆ | $0.43◆ | $0.87◆ | $0.54◆ | 1M◆ |
| DeepSeek V4 Flash (latest)pinned | 52◆ | 69.1◆ | 115.9◆ | $0.14◆ | $0.28◆ | $0.18◆ | 1M◆ |
coding performance
The Intelligence Index and its sub-scores, ranked against every scored model in the catalog. A metric that has not been measured for Kimi K3 has been left empty.
Coding Index
76.2
#5 of 40 scored
Terminal-Bench
85
#4 of 40 scored
Intelligence Index
59.7
#4 of 46 scored
Long-context reasoning
82.7
reasoning across a long context
SciCode
58.7
scientific coding
GPQA Diamond
93.5
graduate-level QA
usage calculator
How far a month of credits goes on Kimi K3.
Input tokensfresh prompt
800Output tokensmodel reply
180Cache read tokensre-read context
50Kcost / request $0.020 · in $3 · out $15 · cache $0.30 per M
fresh input 12%output 13%cache reads 75%
Requests / 30 days
498
$10 credits ÷ $0.020 per request
~100 quick fixes~20 bug fixes~3 feature PRs
what real work costs
Real coding tasks priced end to end on Kimi K3, from a quick lookup to a full-repo agent run.
One agent task
$0.36
180K in at 75% cache hit, 12K out
What you pay
$1.85 /M
all-in across every token that task touched
Sticker input
$3 /M
cache reads bill at $0.30 /M instead
| Task | Tokens in · out | Kimi K3 | Muse Spark 1.2 | Claude Haiku 4.5 |
|---|---|---|---|---|
| Quick lookup / one-liner | 8K · 1K | $0.02 | $0.0063 | $0.0065 |
| Review a 500-line PR | 60K · 4K | $0.13 | $0.05 | $0.04 |
| Fix a bug (agent loop) | 180K · 12K | $0.36 | $0.13 | $0.12 |
| Refactor a module | 320K · 20K | $0.64 | $0.23 | $0.21 |
| Full-repo agent run | 900K · 45K | $1.48 | $0.54 | $0.49 |
frequently asked
What is Kimi K3 best for?+
Kimi K3 is a 2.8T parameter open-weight multimodal reasoning model from Moonshot AI. It is suited for complex coding, knowledge work, and long-horizon agentic workflows, and is particularly strong at navigating large repositories, using tools, debugging, and iterating against images, logs, tests, and runtime feedback. Its architecture uses KDA and Attention Residuals for computational efficiency.
How much does Kimi K3 cost?+
$3/M input and $15/M output, cache reads $0.30/M. In an agent loop most input is cache-read, so the effective input rate is about $1.11/M.Which plan do I need?+
Available on Go and above.
How do I switch to it?+
Run
cmd --model moonshotai/kimi-k3, or type /model in a session and pick it. You can switch mid-session without losing context.Ship code that matches your taste
Command Code is the AI coding agent that continuously learns your taste. Start for $1.
Benchmarks from Artificial Analysis (v4.1)commandcode.ai/models/kimi-k3