Cognitive Computations
1 model built
Cognitive Computations is a community org on Hugging Face, not a commercial AI lab — it's the group behind the Dolphin series of uncensored instruct fine-tunes, built on top of base models from Llama, Mistral, and Qwen. Their approach is to take an existing open-weight base and retrain it to drop refusals and alignment guardrails, giving you a model that answers directly instead of hedging. On OpenKey they have exactly one model listed: dolphin-mistral-24b-venice-edition, a fine-tune of Mistral's 24B base. It's a Dolphin release aimed at users who want an instruction-following model without the built-in content filtering most vendor models ship with.
Lineup
There's no lineup to navigate here — Cognitive Computations has one model on OpenKey, dolphin-mistral-24b-venice-edition, and it's free (both input and output are $0 per 1M tokens). It runs on a 32,768-token context window and handles text-to-text only. If you want an uncensored Mistral-based fine-tune and don't need a large context window or multimodal input, this is the only option from this org. For anything requiring more context or vision/audio support, you'll need a different lab.
- Uncensored (free)FreeJul 9, 2025 33K Free Free ctx 33K · in Free · out Free
Prices per 1M tokens, flat 3% fee included.
Questions
- What does Cognitive Computations charge for their model?
- Nothing. Their only model on OpenKey, dolphin-mistral-24b-venice-edition, is free: $0 per 1M input tokens and $0 per 1M output tokens. There's no OpenKey fee to calculate since 3% of zero is zero.
- Which Cognitive Computations model should I use?
- There's only one: dolphin-mistral-24b-venice-edition, released July 9, 2025. It's a 24B-parameter Dolphin fine-tune of Mistral with a 32,768-token context window, free to use, and built for uncensored instruction-following rather than benchmark performance (no benchmark score is listed for it).
- Is Cognitive Computations a commercial API provider?
- No — it's a community fine-tuning org hosted on Hugging Face, best known for the Dolphin series of uncensored instruct models. They don't run their own API infrastructure; their one model reaches OpenKey through the standard model roster, priced at $0.