OpenChat
Open 7B and 8B models trained with C-RLFT, dormant since May 2024
Overview
OpenChat is a family of open language models led by Guan Wang with Tsinghua University contributors, fine-tuned with C-RLFT, a method for learning from mixed-quality data without preference labels. The 7B OpenChat 3.5 models are Apache 2.0 and the Llama 3-based OpenChat 3.6 carries the Llama 3 licence; there has been no release since May 2024.
Pricing
Pricing shown for reference only. These figures reflect RECATOOLS research as of 14 Oct 2026 and may be out of date or incomplete. This is not financial or purchasing advice — always confirm the current price on the provider’s official website before making any decision.
Use cases
What you can produce with OpenChat
- OpenChat 3.5-0106 weights, a 7B Mistral fine-tune under Apache 2.0
- OpenChat 3.6 8B weights built on Llama 3 under the Llama 3 licence
- The C-RLFT training method and paper (ICLR 2024) for learning from mixed-quality data
- Training code and data format for running C-RLFT on your own conversations
- An OpenAI-compatible serving setup for self-hosting on a single consumer GPU
OpenChat earned attention for C-RLFT, a clean way to learn from mixed-quality instruction data, and for 7B models the project measured as competitive with early-2023 ChatGPT. The weights are still free to download and run on a single consumer GPU.
But the project has been silent since May 2024, issues go unanswered, and the old openchat.team demo domain is reported to serve a malware lure. Keep an existing deployment running if it works; start new projects on an actively maintained small model.
What we found
OpenChat is a family of open language models led by Guan Wang, with contributors from Tsinghua University and sponsorship that included Tsinghua, 01.AI and RunPod. Its contribution is C-RLFT, a fine-tuning method described in a paper presented at ICLR 2024. Most instruction data mixes a little expert material with a lot of mediocre material; C-RLFT treats each data source as a coarse reward label and trains a class-conditioned model, so it learns from both without needing preference rankings.
The models came in quick succession from mid-2023. OpenChat 3.5, a 7B fine-tune of Mistral 7B released in late 2023, was billed in the README as on par with the March 2023 ChatGPT, and the January 2024 update, 3.5-0106, was said to beat xAI's 314B Grok-1 on average and on three of four benchmarks. Those are the project's own measurements. A Gemma-based variant followed in March 2024, and OpenChat 3.6, built on Llama 3 8B, arrived in May 2024 with a claim to outperform Meta's official Llama 3 8B Instruct. The 3.5 models are Apache 2.0; 3.6 carries the Llama 3 licence.
That was the last release. The repository's final commit and its 3.6.1 tag both date from 24 May 2024, 131 issues are open, and an August 2025 issue asking whether the project is dead has no reply. The weights remain downloadable from Hugging Face and the serving code still works for existing users, but there is no first-party API, hosted service or support.
One practical warning. An August 2026 pull request reports that openchat.team, the old demo domain still linked from the README, now sends visitors to a malware lure that tries to trick them into running a PowerShell command. The fix had not been merged in October 2026. Use the GitHub repository and the Hugging Face pages, and treat OpenChat as a well-documented 2023 to 2024 research line rather than a model for new projects.
Compiled by RECATOOLS from public sources: vendor documentation and pricing, code repositories and published reporting.
Notable facts
- OpenChat 3.5 matched GPT-3.5 performance on multiple benchmarks with only 7 billion parameters — an 8x parameter efficiency advantage.
- The C-RLFT training technique was invented in a university lab as a research project, then open-sourced — demonstrating that frontier training techniques can emerge from academic settings.
- OpenChat was one of the first models to demonstrate that learning from negative examples (suboptimal responses) is as important as learning from positive examples.
Frequently asked questions
About this listing
This entry was compiled from publicly available data including OpenChat's official website, press releases, documentation, and reputable third-party publications. RECATOOLS is not affiliated with OpenChat unless explicitly stated.
Third-party AI tools update their pricing, features, availability, and policies frequently. Information here may be outdated by the time you read this — we make reasonable efforts to keep listings current, but cannot guarantee absolute accuracy.
For the latest details, please refer to OpenChat directly →
Spotted something out of date? Suggest an update →
Related reading
Alternatives to OpenChat
More in LLMs & Chat
