Model
Model explorer
Kimi K3 Fast
CLOSEDMoonshot AI · Kimi K3 family · released Jul 27, 2026
Fast-inference variant of Kimi K3, released the same day as K3's open-weight drop (July 27, 2026). Priced 50% above base K3 for lower latency; only listed via Vercel AI Gateway at time of writing, with no separate HuggingFace repo or independently-published benchmark suite distinct from base K3 — left unranked pending that coverage.
ReasoningCodingVisionFunction callingTool useAgentic
—
Elo · unrated
Parameters
2800B
Active params
Undisclosed
Context
1.024M tokens
Architecture
Same 2.8T-parameter Stable LatentMoE architecture as Kimi K3 (16 of 896 experts active per token, Kimi Delta Attention), served on a lower-latency inference stack
License
Proprietary (fast-inference serving variant of the open-weight Kimi K3; no distinct weights published)
Languages
—
API price (in/out)
$4.5 / $22.5
Modalities
text · vision
Benchmark results
Bar shows position within the tracked field; marker = field best
Run it locally
Closed weights — available via API only. No local deployment.
Input / M tok
$4.5
Output / M tok
$22.5
API price $4.5/$22.5 · each benchmark row carries its own source badge (see methodology)