Kimi K3 Arrives on Amazon Bedrock

Kimi K3’s Bedrock launch tests whether Chinese open-weight AI can scale commercially.
30-Second TL;DR
What Changed
Kimi K3 is now available on Amazon Bedrock.
Why It Matters
Distribution through AWS could give Moonshot AI access to international enterprise developers and production workloads. It also gives open-weight model providers a route to monetize through cloud ecosystems rather than direct hosting alone.
What To Do Next
Run a small coding and knowledge-work benchmark comparing Kimi K3 on Bedrock with your current production model.
Key Points
- •Kimi K3 is now available on Amazon Bedrock.
- •AWS positions the model for coding and knowledge work.
- •The deployment tests the commercial model for Chinese open-weight AI.
Deep Insight
Background and context from public sources — not the original article. 10 sources cited.
Enhanced Key Takeaways
- •Moonshot AI's revised commercial license terms require third parties hosting the model as a service to enter a paid agreement once aggregate annual revenue exceeds US$20 million.
- •Kimi K3 is the first open-weight model on Amazon Bedrock to support explicit prompt caching, allowing user-defined breakpoints starting at 1,024 tokens that persist for 30 minutes.
- •The deployment incorporates an enterprise privacy boundary where inference data remains strictly within AWS with zero data retention and is not shared with Moonshot AI.
- •Moonshot AI is actively pursuing global multi-cloud monetization, including seeking up to 30% revenue-sharing deals with major Western cloud providers like Microsoft and Google.
- •The international push aims to bridge a wide revenue gap, as Rhodium Group data indicates the top seven Chinese AI labs together generate only about 10% of OpenAI and Anthropic's combined revenue.
Competitor Analysis
- Developer
- Moonshot AI
- Architecture / Total Parameters
- MoE (2.8T parameters)
- Context Window
- 1,000,000 tokens
- Benchmark Standing
- Top-tier on Artificial Analysis leaderboard, trailing Claude Fable 5 & GPT-5.6 Sol
- Developer
- Anthropic
- Architecture / Total Parameters
- Proprietary Frontier
- Context Window
- Large Context (Proprietary)
- Benchmark Standing
- Frontier leader on Artificial Analysis leaderboard
- Developer
- OpenAI
- Architecture / Total Parameters
- Proprietary Frontier
- Context Window
- Large Context (Proprietary)
- Benchmark Standing
- Frontier leader on Artificial Analysis leaderboard
| Model | Developer | Architecture / Total Parameters | Context Window | Benchmark Standing |
|---|---|---|---|---|
| Kimi K3 | Moonshot AI | MoE (2.8T parameters) | 1,000,000 tokens | Top-tier on Artificial Analysis leaderboard, trailing Claude Fable 5 & GPT-5.6 Sol |
| Claude Fable 5 | Anthropic | Proprietary Frontier | Large Context (Proprietary) | Frontier leader on Artificial Analysis leaderboard |
| GPT-5.6 Sol | OpenAI | Proprietary Frontier | Large Context (Proprietary) | Frontier leader on Artificial Analysis leaderboard |
Technical Deep Dive
- Parameter Scale: Built with a 2.8-trillion total parameter Mixture-of-Experts (MoE) architecture, achieving approximately 2.5× the scaling efficiency of Kimi K2.
- Context Window: Supports a 1-million-token context window designed for continuous reasoning across extensive codebases and enterprise document corpora.
- Multimodal Support: Features native multimodal image-processing capabilities alongside text reasoning.
- Explicit Prompt Caching: First Bedrock open-weight model with explicit caching breakpoints (minimum threshold of 1,024 tokens), retaining cached states for at least 30 minutes to reduce latency and input cost.
- Security & Data Retention: Deployed with AWS enterprise-grade security boundaries featuring zero data retention and isolation from base model training pipelines.
Future ImplicationsAI analysis grounded in cited sources
Timeline
- 2026-07Moonshot AI updates commercial licensing terms with a $20M revenue threshold
- 2026-09AWS announces integration and general availability of Kimi K3 on Amazon Bedrock
Sources (10)
Factual claims are grounded in the sources below. Forward-looking analysis is AI-generated interpretation.
Weekly AI Recap
Read this week's curated digest of top AI events →
AI-curated news aggregator. All content rights belong to original publishers.
Original source: SCMP Technology ↗
This is a summary, not the original. Read the source, or get the weekly briefing.
The weekly digest
One email a week. Unsubscribe anytime.


