Kimi on Fireworks

The Kimi Series models are built by Moonshot AI, an Alibaba-backed company founded by Zhilin Yang. They are known for pioneering open agentic intelligence with state-of-the-art performance in coding, multimodal intelligence, and multi-turn agents. Kimi K3 is their most capable model to-date, released as an open-weights, 2.8 trillion parameter model on July 27th. Kimi K3 is available on all serverless tiers for devs to start building with on Fireworks today.

Get Started Now

Kimi Model Collection

Explore the latest models from Moonshot AI, including the Kimi K2.7, K2.6, and the full Kimi K2 Thinking family. Enjoy state-of-the-art performance in complex agents, coding, and document intelligence.

Moonshot Innovations That Break the Mold

Advanced Reasoning for Agents and Coding

Kimi K3 Achieves SOTA quality in complex multi-agent reasoning and visual tasks, outperforming closed-source models. We ran Kimi K3 against Fable on ~1,000 agentic tasks and Kimi K3 outperformed on security, crypto, and long terminal loops, at up to 50x lower cost than Fable on long loops.

Learn More

Breakthrough Speed

•3X faster inference than closed-source models, delivering up to 201 Token/s
•4.5X faster inference than other providers

Learn more

Full Multimodal Control

Beginning with Kimi K2.5, the Kimi K2+ family of models powers long-horizon agentic workflows with the benefit of Moonshot’s multimodal power. It integrates image and video understanding with transparent 'thinking' modes, ensuring unmatched trust and visibility in every agentic workflow.

Learn more

MoE and Long Context Windows for Superior Agentic Intelligence

The Kimi K2+ Series incorporate a Mixture-of-Experts (MoE) architecture that provides superior efficiency and higher performance. It also provides a large context window that maintains remarkable coherence for complex analysis and multi-turn flows, making it the optimized choice for long-range agentic workflows.

Learn more

Open Models Reach the Frontier

Open models like the Kimi series now deliver production-grade quality while drastically cutting costs. The true competitive advantage belongs to teams that can rapidly adopt these new capabilities and run them reliably, efficiently, and at scale.

Why choose Kimi on Fireworks?

Unmatched Speed That Powers Your Workflows

Consistently rank fastest speed on AA Benchmarks, achieving up to 201 tokens per second with our proprietary performance optimizations.

Learn more

Slash Costs Without Sacrificing Quality

1/10th the Cost of Closed Models: Migrate from expensive closed-source models while keeping output quality and speed high.

Day-Zero Deployments Gets You to Market Faster

Fireworks has been Moonshot AI's Launch Partner for Kimi K2.5, K2.6, K2.7, and now with Kimi K3. As the frequency of model releases increases, Fireworks remains committed to providing day-zero support for devs to start testing and deploying new Kimi models within hours of launch.

CASE STUDY

Genspark Outperforms Closed Models: 12% Better Quality at 50% Lower Cost

Genspark leveraged Fireworks’ Reinforcement Fine-Tuning to outperform state-of-the-art closed-source models in just one month. By training Kimi, they achieved 12% better quality and a 33% increase in tool calls, all while slashing operational costs by 50%. This shift has enabled Genspark to deliver superior, cost-effective AI agents for deep research, documents, and complex web applications.

Read the Case Study

50%

Cost Reduction