Cue Cloud

productmodels.md

Open coding models

Cue Cloud serves open-weight coding models we pull from Hugging Face and run on infrastructure we own: DeepSeek V4, Kimi K2.7 Code, and GLM 5.2.

Open weights. We deploy them.

Cue Cloud serves open-weight coding models we pull from Hugging Face and run on hardware we operate — not a closed frontier API with a Cue sticker. Check the Hub pages yourself.

Flagship models we serve

DeepSeek V4

Open-weight coding flagship

deepseek-v4-pro
Context
1M tokens
Architecture
MoE · 1.6T total / 49B active
Weights
Open (we deploy from Hugging Face)
License
MIT
Best for
Default agent loops, SWE workloads, long-context coding
Source
deepseek-ai/DeepSeek-V4-Pro on Hugging Face →
  • Frontier-class coding scores on open weights
  • Million-token context for repo-scale work
  • Cue Cloud serves the Hub checkpoint on our fleet
request sketch
POST /v1/chat/completions
Authorization: Bearer cue_••••••••
Content-Type: application/json

{
  "model": "deepseek-v4-pro",
  "messages": [
    { "role": "user", "content": "fix the failing test" }
  ]
}

Benchmarks

Frontier-class coding on open weights

Scores for DeepSeek V4 from the upstream model card. Cue Cloud deploys these weights — we are not inventing a private bake-off.

SWE-bench Verified80.6%
DeepSeek V4
80.6%
Claude Opus 4.7
87.6%
Claude Opus 4.8
88.6%
GPT-5.5
82.6%
SWE-bench Pro55.4%
DeepSeek V4
55.4%
Claude Opus 4.7
64.3%
Claude Opus 4.8
69.2%
GPT-5.5
58.6%
MCP-Atlas73.6%
DeepSeek V4
73.6%
Claude Opus 4.7
77.3%
Claude Opus 4.8
82.2%
GPT-5.5
75.3%

DeepSeek scores: vendor-reported Think Max on Hugging Face. Frontier peers: published Opus 4.7 / 4.8 / GPT-5.5 coding comparisons. Not Cue Cloud latency or tok/s.

Verify on Hugging Face →

How Cue Cloud compares to frontier APIs

Job fit, not a fake bake-off. We deploy open Hub checkpoints on our fleet at $1,500/seat/mo · unlimited tokens — not a thin proxy in front of Claude or GPT.

DimensionCue CloudClaude APIGPT API
WeightsOpen Hub checkpoints we deployClosedClosed
Where it runsCue Cloud owned fleetAnthropic APIOpenAI API
LineupDeepSeek V4 · Kimi K2.7 Code · GLM 5.2Opus 4.7 / 4.8 classGPT-5.5 class
Pricing$1,500/seat/mo · unlimited tokensUsage-based · unbounded spendUsage-based · unbounded spend
Seatcue_… key per user on a flat seatOrg / workspace keysOrg / project keys
Lock-inSwap open models on our metalVendor model lockVendor model lock
ProofHugging Face links on every cardClosed weightsClosed weights

Open weights we deploy

These are public Hub checkpoints — not a closed frontier API. Cue Cloud downloads and serves them on hardware we operate. Every model card links to Hugging Face so you can verify the source.

Why not a closed frontier API

Open weights keep the coding loop inspectable and portable. We are not a thin proxy in front of someone else's chat endpoint. Model choice and serving sit with Cue Cloud.

Hardware we run

Inference runs on machines we operate. Capacity, routing, and cost sit with Cue Cloud. The models above decode on that fleet, not on a rented third-party chat API.

Tuned for coding agents

Tool-using turns, repo-scale context, and concurrent agent loops are the normal case. Chat playground demos are not the design target.

Back to explorer
change theme