The fast, filtered one — free, and honest about which it is.
OpenAI · GPT-4o family · Not disclosed parameters · 128K context · Free · Filtered by its provider
OpenRogue is not anti-frontier. It is anti-refusal, which is a different and much narrower complaint, and GPT-4o Mini is the clearest illustration of the distinction. It is a genuinely good small model: quick, cheap enough to give away, careful with formatting, comfortable with 128K of context, and better than most open models of any size at the boring work that fills an actual day — summarising a thread, cleaning up a paragraph, converting notes into a table, explaining an error message, drafting the email you have been avoiding. It is on OpenRogue's free tier for exactly those jobs, because pretending they do not exist would be posturing.
What it also comes with is OpenAI's content policy, trained in and non-negotiable. Ask it for blunt clinical detail, a villain who wins, security internals, or anything adult, and you will get the apology paragraph — that is the design, not a defect, and no platform can remove it. The difference on OpenRogue is what happens next: instead of rephrasing your question five times to sneak past a filter, you change the model in the dropdown and keep the same conversation. Dolphin Mistral 24B Venice is also on the free tier and does not have the reflex. Treat GPT-4o Mini as the fast first pass and the uncensored lineup as what it hands off to.
Yes — it sits on the free tier, so you can use it without a card. The free tier has a daily message cap shared across every free model; a paid plan removes the cap and opens the larger uncensored and frontier models.
Yes. OpenAI trains its content policy into the model itself, so it refuses adult material, softens dark fiction and hedges on a wide band of legal-but-sensitive topics. OpenRogue adds nothing on top of that, but it also cannot take it away — the behaviour lives in the weights.
Because of what sits next to it. On OpenRogue, GPT-4o Mini shares one dropdown and one conversation history with Dolphin Venice, Hermes and the rest of the uncensored lineup, so the moment it refuses you switch model rather than switching product. See the full head-to-head on the ChatGPT comparison page.
Ordinary throughput: summarising, reformatting, explaining, drafting, light coding help, and anything where speed matters more than depth. It is a strong default for the parts of your day that are not controversial at all — which, honestly, is most of them. Where it gets thin is long-range reasoning and creative voice; a 70B from the lineup will beat it on both, and the free Dolphin Venice will beat it on bluntness.
OpenAI has not published a parameter count for it, so neither will we. What is documented is a 128,000-token context window and latency well below the full GPT-4o. If parameter counts matter to you, the open models in the lineup publish theirs — from Lunaris at 8B up to Hermes 3 at 405B.
Dolphin Mistral 24B Venice Edition. It is also on the free tier, it is uncensored at the fine-tune level rather than un-prompted, and at 24B it is the strongest free option for direct answers. MythoMax and Lunaris cover the creative side.
Related models: Dolphin Venice · Lunaris · MythoMax