Moonshot’s Kimi K3 lands in Japan with 2.8T open weights and $3/$13 pricing
DavidBennett__ · x · 2026-07-28
Moonshot’s Kimi K3 launches in Japan with frontier-level open weights and lower pricing
ai& says it has begun offering Kimi K3 in Japan through its ai& Inference service. The post frames K3 as a 2.8T-parameter open-weight model that closes much of the gap to closed frontier models.
- Benchmark claim: K3 scores 57 on the Artificial Analysis Intelligence Index, just behind Claude Fable 5 (60) and GPT-5.6 Sol (59), while exceeding other listed models such as Claude Opus 4.8.
- Use cases: Moonshot says K3 leads open-weight models on agentic search, reasoning, coding, and document understanding.
- Pricing: $3 input / $13 output per 1M tokens, which the post says is far below Claude Opus 5.
- Deployment: ai& says K3 is running day one across NVIDIA and AMD GPUs via vLLM, SGLang, and its own software stack, with heterogeneous operation and domestic processing in Japan.
The announcement also emphasizes that K3 is now available via inference API for self-hosting and verification.
Related event: Moonshot AI's Kimi K3 Launches in Japan(2 posts)→
More from Infra
- Is inference latency becoming the biggest bottleneck for production AI agents? — Euphoric_Sea632 · 2026-09-11
- LLM Serving Metrics Thread: Why TPOT and Uptime Make or Break User Experience — abhijithneil · 2026-09-11
- PlanetScale launches sharded Postgres: 768 servers acting as one, 1PB scale — dhruv2038 · 2026-09-11
- Can a 7900 XTX 24GB run Qwen locally? Reddit seeks ROCm tok/s benchmarks — thenomadexplorerlife · 2026-09-11
- RTK Terminal Compression Cuts Tokens but Leaves Your AI Coding Bill Unchanged — Bartaseth · 2026-09-11
- SF Compute founder: buying compute is 'an absolutely awful experience' right now — IgorCarron · 2026-09-11