Kimi K2 Instruct

Get started with this model in 10 lines of code! The model ID is moonshotai/Kimi-K2-Instruct and the pricing is $1 for input tokens and $3 for output tokens.

    
      from together import Together

      client = Together()
      resp = client.chat.completions.create(
          model="moonshotai/Kimi-K2-Instruct",
          messages=[{"role":"user","content":"Code a hacker news clone"}],
          stream=True,
      )
      for tok in resp:
          print(tok.choices[0].delta.content, end="", flush=True)
    

Architecture Overview:

• ‍‍1 T-parameter MoE with 32 B activated parameters
• Hybrid MoE sparsity for compute efficiency
• 128K token context for deep document comprehension
• Agentic design with native tool usage & CLI integration‍

Training Methodology:

• ‍‍Pre-trained on 15.5 T tokens using MuonClip optimizer for stability
• Zero-instability training at large scale

‍Performance Characteristics:‍

• SOTA on LiveCodeBench v6, AIME 2025, MMLU-Redux, and SWE-bench (agentic)

• Use natural language instructions or tool commands
• Temperature ≈ 0.6: Calibrated to Kimi‑K2‑Instruct’s RLHF alignment curve; higher values yield verbosity.
• Kimi K2 autonomously invokes tools to fulfill tasks: Pass a JSON schema in tools=[…]; set tool_choice="auto". Kimi decides when/what to call.
• Supports multi-turn dialogues & chained workflows: Because the model is “agentic”, give a high‑level objective (“Analyse this CSV and write a report”), letting it orchestrate sub‑tasks.
• Chunk very long contexts: 128 K is huge, but response speed drops on >100 K inputs; supply a short executive brief in the final user message to focus the model.

Kimi K2 shines in scenarios requiring autonomous problem-solving – specifically with coding & tool use:

• Agentic Workflows: Automate multi-step tasks like booking flights, research, or data analysis using tools/APIs
• Coding & Debugging: Solve software engineering tasks (e.g., SWE-bench), generate patches, or debug code
• Research & Report Generation: Summarize technical documents, analyze trends, or draft reports using long-context capabilities
• STEM Problem-Solving: Tackle advanced math (AIME, MATH), logic puzzles (ZebraLogic), or scientific reasoning
• Tool Integration: Build AI agents that interact with APIs (e.g., weather data, databases).