Red Router

This account is verified. Learn more

@redrouterorg

Abliterated and Open models for offensive cybersecurity and reverse engineering. Zero Data Retention.

New York, New York redrouter.org Joined October 2026

96 Following 259 Followers 1.1K Posts

Latest posts 5

Pinned post

Red Router @redrouterorg 16h

Red Router Abliterated Benchmarks are live!

This is the first and only continous measurement of Speed, latency, reliability, and listed price across providers. 🧵(1/3) benchmarks.redrouter.org

0:12

Red Router @redrouterorg 1h

keep the incredible ideas coming! some awesome new benchmarks planned too 🤫

Adolan Verified account @adolandev 14h

@redrouterorg Went through the methodology. Love that failed requests stay in the denominator and suppliers are pinned where possible.

Would be cool to see concurrency tests next. Single-request tok/s is nice, but p90 latency with 8 parallel agent calls tells you much more about a provider!
View quoted post on X

Red Router @redrouterorg 2h

firmware that knows which weights are which needs a hash list nobody maintains lol

Red Router @redrouterorg 2h

fable 5.2 dropped today and the pile of ai written bug reports is about to get taller

bugcrowd's oct 8 throttling update: hit 10 submissions in 90 days on managed programs and your accuracy gets checked. under 50% and you're throttled. n/a counts as a rejection. throttled accounts get six submissions a week, and new accounts start throttled

verify the evidence before the model writes the report:
redrouter.org/blog/hackerone…

redrouter.org AI on HackerOne and Bugcrowd: a safe workflow Use AI in HackerOne and Bugcrowd workflows without guessing the rules. Check program scope, data handling, evidence and submission quality first.

Red Router @redrouterorg 2h

fable 5.2 day, so half the tl is editing claude code env vars tonight

if you point claude code at another backend, set every tier, not just ANTHROPIC_MODEL. sonnet, opus, haiku, fable and CLAUDE_CODE_SUBAGENT_MODEL all get the same id, or subagents can fall back to a model you aren't paying for there

on redrouter that's glm-5.3-rr or glm-5.3-flash-rr. known gaps: images and pdfs return 400, effort settings get ignored, 8k output cap per response

full config:
redrouter.org/blog/claude-co…

redrouter.org Use low-refusal models in Claude Code, OpenCode and Cline Point Claude Code at RedRouter's Anthropic Messages endpoint, or OpenCode and Cline at its OpenAI-compatible API. Configs, model choice and known gaps.