GLM-5.2 is live. #1 throughput on OpenRouter, pay-per-token on FriendliAI. Try it today ➜
Blackfrost-AI
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Model Details
Model Provider
Blackfrost-AI
Model Tree
Qwen/Qwen3.8-27B
Input Modalities
Output Modalities
Supported Functionality
| Property | Value |
|---|---|
| Upstream model | Qwen/Qwen3.8-27B |
| Architecture | Qwen3_5ForConditionalGeneration |
| Weight format | BF16 Transformers checkpoint |
| Quantization | None |
| Maximum configured context | 262,144 tokens |
This card intentionally excludes internal transformation, training, data, deployment, and reproducibility details.
| Dataset | Prompts | Substring refusal flags | Flag rate | Manually confirmed safety refusals |
|---|---|---|---|---|
| AdvBench | 150 | 3 | 2.0% | 0 |
| StrongREJECT | 150 | 0 | 0.0% | 0 |
| XSTest | 150 | 0 | 0.0% | 0 |
Harmful subset (AdvBench + StrongREJECT): 300 prompts, 3 substring flags (1.0%), 0 manually confirmed safety refusals, and 1 strict task non-completion (0.33%). Manual review classified the other 2 flags as capability disclaimers followed by substantive responses.
Quality checks: 0 API errors, 0 empty responses, and 0 manually confirmed incoherent responses. The automatic detector produced 17 incoherence flags, all caused by repeated Markdown or diagram separator characters rather than degraded output.
This checkpoint is intended for private, controlled research and evaluation. It may produce inaccurate, biased, unsafe, or otherwise undesirable output. The checkpoint name is not a safety guarantee or a claim of suitability for deployment. Evaluate it for the intended domain and apply appropriate access controls and safeguards before any broader use.
The upstream model is released under the Apache License 2.0. Review the upstream Qwen3.8-27B model card for its documentation, license information, and original limitations.
| 450 |
| 3 |
| 0.7% |
| 0 |