Model
- Model: Kimi K2.7 Code
- Provider-facing model ID:
moonshotai/Kimi-K2.7-Code
- Upstream/provider: Moonshot AI model served through Together AI serverless
- API family: Together AI OpenAI-compatible chat/completions serverless inference
- Release date: Not stated on the Together serverless model catalog page checked on 2026-07-04.
Evidence
Capabilities and limits
- Context window: 262,144 tokens.
- Coding-oriented model name: Kimi K2.7 Code.
- Quantization / serving tier shown by Together: FP4.
- Catalog flags: the checked table row marks the final two feature columns as
Yes / Yes.
Pricing
Together catalog row checked on 2026-07-04 lists:
- Input: /bin/bash.95 / 1M tokens.
- Cached input: /bin/bash.19 / 1M tokens.
- Output: .00 / 1M tokens.
Cross-provider availability
- Confirmed in this check: Together AI serverless.
- Not confirmed in this check: Azure, Amazon Bedrock, Google Vertex.
- Related catalog signal: existing Atlas agent-stack records mention GitHub Copilot CLI support for
kimi-k2.7-code, but the compute model graph does not currently contain a Kimi K2.7 Code model node.
Graph work needed
- Add or update a compute model record for Kimi K2.7 Code if Atlas decides to track Moonshot/Kimi as first-class model versions.
- Add provider availability edge/claim for
provider:together-ai with provider model ID moonshotai/Kimi-K2.7-Code.
- Add evidence source and claims for context window, pricing, API family, and Together serverless availability.
- Consider whether a
model-family:kimi-k2 / Moonshot provider record is needed, or whether this should remain an inference-provider catalog claim only.
Transport-adapter proxy
Together AI uses an OpenAI-compatible API surface, so the existing OpenAI-compatible transport/proxy path should be the expected integration shape. No Anthropic Messages, Gemini GenerateContent, Bedrock Converse, or Cohere-native adapter support appears required for Together-hosted use.
Model
moonshotai/Kimi-K2.7-CodeEvidence
moonshotai/Kimi-K2.7-Code.Capabilities and limits
Yes/Yes.Pricing
Together catalog row checked on 2026-07-04 lists:
Cross-provider availability
kimi-k2.7-code, but the compute model graph does not currently contain a Kimi K2.7 Code model node.Graph work needed
provider:together-aiwith provider model IDmoonshotai/Kimi-K2.7-Code.model-family:kimi-k2/ Moonshot provider record is needed, or whether this should remain an inference-provider catalog claim only.Transport-adapter proxy
Together AI uses an OpenAI-compatible API surface, so the existing OpenAI-compatible transport/proxy path should be the expected integration shape. No Anthropic Messages, Gemini GenerateContent, Bedrock Converse, or Cohere-native adapter support appears required for Together-hosted use.