OpenAI's multimodal workhorse, filter included.
OpenAI · GPT-4o · Not disclosed parameters · 128K context · Pro · $20/mo · Filtered by its provider
GPT-4o is the model that made multimodal feel ordinary. Hand it a screenshot, a photograph of a whiteboard, a scanned invoice, a chart with no underlying data, a page of handwriting — and it reads the thing rather than describing it. Paired with OpenRogue's vision and file upload, that makes it the natural pick for any task that starts with an image or a document you cannot be bothered to retype. It is also unusually good at tone: it holds a register, matches a house style, and produces professional writing that needs less editing than most models of any size. For work correspondence, client-facing drafts and anything where the output has a reader with expectations, it earns its place.
The trade is the same one every OpenAI model carries, and it is worth stating flatly on its own page: GPT-4o enforces OpenAI's content policy from inside the weights. Adult material is out. Dark fiction gets steered toward redemption. A surprising amount of ordinary research — pharmacology, security, criminal procedure, anything with an edge — comes back wrapped in a disclaimer or not at all. On OpenRogue that is a routing problem rather than a dead end. Read the image with GPT-4o, then move the same conversation to Hermes 4 70B for analysis it will not hedge on, or Magnum v4 72B for prose it will not sand down. The context comes with you when you switch.
Yes — vision and file upload are part of the product, and GPT-4o is the strongest model in the lineup for reading a screenshot, a photo of a document or a chart with no data behind it. Upload the file, ask your question, and switch models afterwards if the follow-up runs into policy.
No. OpenAI trains its content policy into the model, so GPT-4o refuses adult content, softens dark or graphic fiction, and hedges on a wide band of security, medical and legal-adjacent questions. OpenRogue layers nothing extra on top, but the behaviour is in the weights and cannot be turned off by a platform.
Mini for volume and speed on ordinary work — it is on the free tier. Full GPT-4o for anything visual, anything long, and anything where the writing quality is the deliverable. Both carry the same content policy, so neither is the model you switch to when you hit a refusal.
Depends what it refused. For analysis and research, Hermes 4 70B reasons hard without the reflex. For fiction and roleplay, Magnum v4 72B was trained toward frontier-grade prose without the fence. For a blunt direct answer, Dolphin Mistral 24B Venice. All three are in the same dropdown.
OpenAI has never published a parameter count, so we do not quote one. The published figures are a 128,000-token context window and native multimodal input. The open models in the lineup do publish their sizes, which is one reason people who care about that pick them.
The model is the model, but the product around it is not. ChatGPT wraps it in one vendor's ecosystem and one vendor's policy end to end; here it is one option among many, sharing a conversation with models that answer what it will not. The comparison page covers the differences in detail.
Related models: Hermes 4 · WizardLM-2 · GPT-4o Mini