
TTFT on Kimi K3.1 Pro Abliterated should also be better now.
More capacity has been added for Kimi K3.1 Pro Abliterated TPS should be way better now although prefill speed is still being worked on

Abliterated models available via OpenAI-compatible API. No KYC and no gated access.

TTFT on Kimi K3.1 Pro Abliterated should also be better now.
More capacity has been added for Kimi K3.1 Pro Abliterated TPS should be way better now although prefill speed is still being worked on

we really should make our own cli harness with easy login and some cybersecurity tool integrated. any advice on what we could fork? perhaps opencode, in a way that keeps the native updates working

As usual, you can expect the price to be lowered in the next days. I mean do I even need to tell you? 😂 lowering prices here is routine

More capacity has been added for Kimi K3.1 Pro Abliterated TPS should be way better now although prefill speed is still being worked on

This is the first model we abliterated in-house, with help from a trusted partner. It’s still a Preview, so expect improvements along the way. Currently running at ~100 TPS average, 65–140 depending on load.

⚡️ Introducing Kimi K3.1 Pro Abliterated, the most powerful open model for cyber and coding. • Native MXFP4 • 1M context • Zero data retention $2.5 input / $0.25 cache / $5 output

Big things coming Stay tuned ☺️

Promo's over

🤯 Cache rate for Abliterated DeepSeek V4.1 reduced from $0.025 to $0.019 We also offer a new tier: DeepSeek V4.1 Flash Abliterated Fast which runs at the previous cache rate $0.025 and has faster speeds.

Any small abliterated models you'd like to see on Refuseless?

DeepSeek V4.1 Flash 1017 Openrouter $0.038 in / $0.50 out Refuseless $0.01 in / $0.08 out

Abliterated DeepSeek V4.1 Flash 1017 is again FREE for new users for a limited time. • 1M context window • Native FP8 • Zero data retention Come try it on refuseless.com, just sign up and create your api key! Limited capacity on free tier, first come first serve.

Tested a cyber post trained version of Qwen3.8 27B and unfortunately, it's not good enough to be added on Refuseless. Here's how DeepSeek evaluated the answers of Qwen on a complex cybersecurity repo.
Would you use abliterated Qwen3.8 27b if we hosted it at a very low price?

DeepSeek V4.1 Flash Abliterated price discount just became permanent

GLM 5.3 Flash Abliterated cache price reduced to $0.025

Would you use abliterated Qwen3.8 27b if we hosted it at a very low price?