NEO
NEO = Neural Expert Orchestrator.
Release: v1. This repository contains a model card, not NEO model weights or a trained adapter.
NEO 1 is the HYPERneocloud serving name for DeepSeek-V4-Flash-0731 through Cloudflare Workers AI, using the explicit upstream identifier @cf/deepseek-ai/deepseek-v4-flash-0731.
The public upstream weight revision is 7872f01b1d1fe23eabc4c98b48bffcef5a386062. This identifies the upstream artifact for provenance; it does not prove Cloudflare's managed runtime loads that exact Hugging Face commit. No NEO fine-tuned adapter is claimed.
Upstream model and license:
- https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731
- https://developers.cloudflare.com/workers-ai/models/deepseek-v4-flash-0731/
- Upstream license: MIT. Preserve the original license and attribution in redistributed artifacts.
The serving API keeps final answer text in content and reasoning in the separate reasoning_content field. Tool clients may preserve that separate field for follow-up turns; applications should display only content as the final answer. Streaming is buffered until completion and fails closed on unverified output boundaries.
NEO price: $0.68 per million input tokens and $1.88 per million output tokens. Customer keys, checkout and customer launch remain subject to release verification. Private Workers AI greeting and tool-loop smoke tests are not a capability benchmark. Production route bytes and bindings are verified; an authenticated production NEO request and Cursor end-to-end compatibility remain unverified.
The earlier R1-Distill-Qwen-14B self-hosted configuration is a prototype lane, not NEO 1's current serving base. No inference or automatic fallback to that prototype is part of this release.