1 source 50m ago
OrcaCyber Releases Abliterated GLM-5.3 Flash Weights For Security Research
OrcaCyber, the cybersecurity team at the OrcaRouter inference gateway, released weights for an abliterated version of Z.ai's open-weight GLM-5.3 Flash on October 27, 2026.
Reporting from 1 source: ASCII.jp.
Why it matters
OrcaCyber is turning abliteration into a standing product line, adding verified models to its Cyber Tier one at a time so defenders can attack-test their own infrastructure with a model that will not refuse the request.
Key Facts
- OrcaCyber released abliterated weights for Z.ai's GLM-5.3 Flash on October 27, 2026.
- The model is free to use for 48 hours after the announcement through the Orca Cyber Tier API.
- OrcaCyber lists CyberBench 67.9 and SWE-bench 92.0 among the developer-announced benchmarks.
- FlashLabs provides OrcaRouter, the inference gateway, for the Japanese market.
The model relaxes refusal responses for security research and is offered free for 48 hours through the Orca Cyber Tier API, alongside OrcaCyber Zero and OrcaCyber Ultra. OrcaRouter is provided in Japan by FlashLabs.
Abliteration here is done in the weights, not through jailbreak prompts or extra training. OrcaCyber describes the method as removing the directional component that produces refusal responses, so the relaxed filtering travels with the model file itself. The release sits under a stated policy of defending with every possible model: because different models surface different vulnerabilities, the team abliterates each verified model it approves and adds it to the tier. GLM-5.3 Flash is a mixture-of-experts model with 320 billion total parameters and 18 billion active, and Z.ai positions it as an open-weight frontier model. OrcaRouter's Cyber Tier carries Zero Data Retention and application-based access.
Written with AI by Yomimono from the 1 cited source below, including Japanese-language reporting where cited. See our editorial policy.
Add Yomimono as a preferred source on Google