MoonshotAI iconMoonshotAI: Kimi K2 0905 (exacto)

Model Type

Proprietary model icon

Proprietary Model

API access only

Recommended Use Cases

Text Generation

Try Kimi K2 0905 (exacto)

Kimi K2 0905 (exacto) is an OpenRouter routing variant of Kimi K2 0905 that prioritizes providers with verified high-accuracy tool calling, offering more reliable agentic performance at a modest cost premium.

The exacto endpoints are specifically focused on tool-calling, and should not be viewed as a broader statement on endpoint or provider quality.

  • OpenRouter

Overview

Kimi K2 0905 (exacto) is not a separate model from Moonshot AI, but rather OpenRouter's curated routing endpoint that directs requests only to providers that have demonstrated superior tool-calling accuracy. This addresses the significant variance in tool-calling reliability observed across different inference providers.

Key Difference from Standard K2 0905

The standard Kimi K2 0905 endpoint routes to any available provider, which can result in inconsistent tool-calling behavior. The exacto variant:

  • Routes only to providers verified for tool-calling accuracy
  • Reduces tool-calling failures measurably
  • Improves tool utilization reliability
  • Costs approximately 30-50% more than standard routing

Why Exacto Exists

OpenRouter observed significant provider variance in:

  • Tool-calling propensity (how often models use available tools)
  • Tool-calling accuracy (correct function calls and parameters)
  • Reliability across different tool schemas

Benchmarking on tau2-Bench and LiveMCPBench showed material improvement when routing through exacto-qualified providers.

Technical Specifications

SpecificationValue
Base ModelKimi K2 Instruct 0905
Total Parameters1T
Active Parameters32B
Context Length256K tokens
RoutingExacto-qualified providers only

When to Use Exacto

Use exacto when:

  • Building agentic workflows with critical tool calls
  • Reliability matters more than cost
  • Experiencing tool-calling failures with standard endpoint
  • Running production workloads requiring consistent behavior

Use standard K2 0905 when:

  • Cost optimization is priority
  • Tool calling is not critical to workflow
  • Testing or development environments
  • Simple chat/completion tasks

Availability

  • OpenRouter: moonshotai/kimi-k2-0905:exacto
  • Same underlying model: Identical capabilities to standard K2 0905
  • Provider selection: OpenRouter manages qualified provider pool

Role in Series

This is a routing variant, not a distinct model:

  • Kimi K2 0905: Standard routing, lower cost
  • Kimi K2 0905 (exacto): Verified tool-calling providers, higher reliability (this variant)

Both use the same Kimi K2 Instruct 0905 model weights.

Links