Kimi K3.1 Pro Highspeed hits Amazon Bedrock with 2-3x faster output and 1M-token context
AWS ML Blog · rss · 2026-11-11
Moonshot AI's Kimi K3.1 Pro Highspeed is now on Amazon Bedrock. Built on the same 3T-class open-weight architecture as Kimi K3 with a 1M-token context window, it adds substantially more post-training compute targeting long-horizon agentic work, fewer factual errors, professional tasks like financial modeling, and coding for game engines, 3D scenes, and CAD scripts — at 2-3x the output speed of Kimi K3.1 Pro.
Integration
- Works via OpenAI-compatible Chat Completions/Responses APIs plus Bedrock Converse/Invoke; existing Kimi K3 users only change the model ID.
- The global.moonshotai.kimi-k3.1 inference profile is cheapest; us/in profiles serve data-residency needs.
- The model always reasons first; reasoningeffort (low/high/max, default max) directly cuts latency and tokens on simple tasks.
Cost levers
- Flex tier serves requests at half Standard price in exchange for higher latency — suited to evals, summarization, and background agent tasks; tiers can be mixed per request.
- Explicit prompt caching: mark stable prefixes (≥256 tokens) with promptcachebreakpoint; caches last ≥30 minutes, reads are discounted and don't count against TPM quotas (not available on Converse).
Data stays within the AWS boundary, isn't shared with Moonshot AI or used for training, with zero retention and zero operator access by default.
More from Infra
- Rumor: DeepSeek to pretrain ~2T-param V5 on Huawei Ascend 950, release may slip to 2027 — teortaxesTex · 2026-11-11
- WSJ: Masayoshi Son Over-Reserved a Big Chunk of the Gas Turbine Market in AI Frenzy — mattwbaker · 2026-11-11
- FT: Commerce Dept stalls Megaspeed export blacklist despite 4 agencies pushing over Nvidia chip smuggling — rohanpaul_ai · 2026-11-11
- Run Pi Coding Agent Locally with TensorSharp + Qwen3.8-27B: Full CLI & SDK Tutorial — fuzhongkai · 2026-11-11
- Spotify's viral Claude Code token trick tested: the real ~90% savings, numbers and prompt inside — aakashgupta · 2026-11-11
- One Opus run cost $1.99 — 94% went to reading files, not writing code — aakashgupta · 2026-11-11