MoonshotAI/Kimi-k1.5 resource

Health v2 · maintenance only

Flags: no_releases no_license

Adoption not part of the score

3471 stars · 236 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-29, confidence not recorded

The official repository for Kimi k1.5, Moonshot AI's o1-level multimodal LLM trained with scaled reinforcement learning, primarily containing the technical report and benchmarks rather than runnable training code. It describes long-context RL (up to 128k), online mirror descent policy optimization, and joint text-vision reasoning.

Use cases

• understand how to scale reinforcement learning with LLMs
• learn long-CoT RL training techniques for reasoning models
• compare o1-level reasoning model benchmarks
• study multimodal reasoning model design
• research policy optimization methods like online mirror descent for LLMs
• find a reference for training reasoning without MCTS or process reward models

When to choose

• you want the technical details and benchmark results of the Kimi k1.5 reasoning model
• you are researching RL-based long chain-of-thought training for LLMs
• you need a citation for scaled RL with multimodal LLMs

When to avoid

• you need runnable training or inference code
• you want a ready-to-use model weights release or API
• you need a production LLM tool rather than a research report

Facets

learning-resource · maturity experimental

llm-training reinforcement-learning machine-learning large-language-models reinforcement-learning artificial-intelligence deep-learning python llm reasoning chain-of-thought multimodal research-report policy-optimization long-context gpu linux

Member repositories

For agents

markdown · JSON · MCP: product_card(name="MoonshotAI/Kimi-k1.5")