Kimi K3 lands on Amazon Bedrock: first 2.8T-param open-weight model with explicit prompt caching
AWS ML Blog · rss · 2026-09-19
Moonshot AI's Kimi K3 is now available on Amazon Bedrock — its most capable model and the first open-weight model to reach 2.8 trillion parameters.
- Native vision plus a 1M-token context window, with 2.5x scaling-efficiency gains over Kimi K2
- First open-weight model on Bedrock to support explicit prompt caching: mark reusable prefixes (≥1,024 tokens), writes billed at a premium but kept ≥30 minutes; cache hits get discounted input tokens and skip rate-limit quotas
- Invoked via cross-region profiles (global.moonshotai.kimi-k3); global routing costs 10% less than geographic profiles
- Data stays inside the AWS boundary with zero retention and zero operator access
- Full setup examples for the OpenAI SDK, OpenCode coding assistant, and Hermes Agent, including a demo building a single-file browser game over long-horizon tasks
- Bedrock has added dozens of open-weight models since 2025 from DeepSeek, Google, MiniMax, Mistral, NVIDIA, OpenAI, Qwen and others
More from coding & agent
- Dev launches Jev Search: free open-source tool that picks where to search and ranks results — gaganghotra_ · 2026-09-19
- Ex-Meta Llama 3 RL lead joins Merrai, an AI memory-layer startup, as advisor — misovalko · 2026-09-19
- Braintrust adds Jev as a judge scorer: typed decisions at up to 193.6× speed and 444.6× lower cost — multiply_matrix · 2026-09-19
- Chrome team publishes a framework for designing WebMCP tools for agentic workflows — gaganghotra_ · 2026-09-19
- I spent $3.40 on Jev in 24 hours: it will be Jev + LLMs, not Jev vs LLMs — gaganghotra_ · 2026-09-19
- Agentic Benchmark Checklist paper shows flawed agent benchmarks skew results by up to 100% — ddkang · 2026-09-19