Moonshot AI releases Kimi K3, a 2.8 trillion parameter open-weight model achieving frontier-class performance
Open source release ● Confirmed 99% confidence first seen
Moonshot AI released Kimi K3, a 2.8 trillion parameter open-weight language model with 1 million token context window on July 27, 2026. The model achieved top-tier benchmark rankings, including #1 on Arena's frontend coding leaderboard and a score of 57 on Artificial Analysis's Intelligence Index, with pricing set at $3 per million input tokens and $15 per million output tokens. The weights were made publicly available on Hugging Face, enabling developers to self-host the model, though initial API subscriptions closed within 48 hours due to overwhelming demand.
Decision brief
- What changed
- Moonshot AI released Kimi K3, a 2.8 trillion parameter open-weight language model with a 1M token context window, on July 27, 2026, achieving top-tier benchmark results (including #1 on Arena's frontend coding leaderboard and 57 on Artificial Analysis's Intelligence Index) at $3/$15 per million input/output tokens; weights were published on Hugging Face after initial API subscriptions closed within 48 hours due to demand.
- Why it matters
- This is the largest open-weight model to date reaching near-frontier performance, narrowing the open-vs-closed capability gap to an estimated 4-7 months and giving enterprises a self-hostable, lower-cost alternative to proprietary US models for coding and agentic workloads. It intensifies pricing and competitive pressure on Anthropic, OpenAI, and cloud AI vendors, triggered market anxiety ('spooked Wall Street'), and reignited US policy debate over restricting Chinese open models, with major labs reportedly lobbying regulators while critics call this protectionist.
- Evidence
- Coverage is extensive and consistent across 33 independent sources (Latent Space, The New Stack, Interconnects, TechCrunch, The Verge, AWS, Together AI, Zvi, Import AI) corroborating parameter count, benchmark rankings, pricing, and the 48-hour subscription closure; multiple outlets independently confirm the open-weight release and infrastructure requirements (AWS deployment guides, Together AI hosting).
- What remains uncertain
- It's unclear whether Kimi K3's benchmark performance translates to real-world reliability versus overfitting to benchmarks (Zvi notes a 4-6 month capability gap when adjusted); the extent of distillation from Claude versus independent architectural gains is disputed among sources; government/enterprise adoption for sovereign AI use cases remains speculative, and the policy outcome on potential US restrictions on Chinese open models is unresolved.
- Monitor next
- Watch whether the US administration moves forward with layered restrictions on Chinese open-weight models, and whether Moonshot resumes/expands API subscription capacity.
Analytical support, not advice — assumptions and open questions stated above.
Source coverage
[AINews] Kimi K3 2.8T-A50B: the largest open model ever released; Opus 4.8-class at Sonnet 5 pricing
Latent Space
Kimi K3 tops Arena’s coding leaderboard — and it’s open-weight
The New Stack
[AINews] not much happened today
Latent Space
The Sequence Radar #897: Last Week in AI: China, Compression and the Open-Model Race
TheSequence
Kimi K3 Pushed Open Models Toward the Frontier
The Neuron
newsletter
Import AI 465: Open vs closed gaps; Kimi K3; Demis' big policy plan
Import AI
📈 Data to start your week
Exponential View
On Kimi K3: Its Capabilities And Related Discontents
Zvi (Don't Worry About the Vase)
Kimi K3: The open-weights escalation
Interconnects
[AINews] not much happened today
Latent Space
Moonshot launched Kimi K3. Then demand shut down subscriptions in 48 hours.
The New Stack
Open models recap: more on Kimi K3, Qwen 3.8, Xi's WAIC speech, distillation, the open-closed gap, and what's next
Interconnects
Kimi and Xi
ChinaTalk
🔮 Will Kimi K3 change the economics of AI?
Exponential View
Kimi K3 vs Claude Fable 5 on DeepSWE: Cost and Coding
Together AI
‘AI communism’, rogue models, and the why Kimi K3 spooked Wall Street
TechCrunch AI
independent
Kimi K3 vs GPT-5.6 Sol on DeepSWE: Cost, Coding, and Routing
Together AI
Making sense of the panic over Chinese AI
TechCrunch AI
independent
Silicon Valley Splits Over Closing the Borders to Chinese AI
TLDR
newsletter
Why China is giving away its best AI models
The Verge
independent
Moonshot opens Kimi K3 weights — but few can run it
The New Stack
Kimi AI and kvcache-ai Open Sources ‘AgentENV’: A Distributed System that Powers Agentic Reinforcement Learning (RL) Training for Kimi K3
MarkTechPost
moonshotai/Kimi-K3
Simon Willison
Moonshot Releases Kimi K3 Model Weights on Hugging Face
The Neuron
newsletter
Building Non-Interactive Agentic Coding Workflows with Moonshot AI’s Kimi CLI, JSONL Streaming, Testing, and Session Memory
MarkTechPost
Together AI announces strategic partnership with Moonshot AI to natively serve Kimi models
Together AI
Moonshot AI Open-Sources MoonEP: A Perfectly Balanced Expert Parallelism Library for MoE Training
MarkTechPost
With Moonshot’s free Kimi K3, China changes the sovereign AI playbook
Rest of World
Deploying Kimi K3 on AWS
AWS Machine Learning
Deploying Kimi K3 on Amazon SageMaker HyperPod and Amazon EKS
AWS Machine Learning
Kimi K3: The Complete Developer Guide
Together AI
AMD Releases Instella-MoE-16B-A3B: A Fully Open Mixture-of-Experts LLM With 2.8B Active Parameters Trained On Instinct GPUs
MarkTechPost
Thinking Machines Lab Releases Inkling-Small: A 276B Total, 12B Active Open Weights Multimodal MoE Model
MarkTechPost