What Developers Need to Know About GPT‑4’s New 8K Context and Multimodal Capabilities

OpenAI announced that all paying API users can now access GPT‑4 with an 8K‑token context window (up to 32K tokens) and that the company will open API access to new developers by the end of the month.

The model, released in March 2023, accepts both text and image inputs, achieves human‑level performance on many benchmarks, and is trained on publicly available data plus licensed sources.

Key Features

Creativity : Generates, edits and iterates on creative tasks such as songwriting, script writing, and style‑adapted content.

Visual Input : Accepts images and can produce descriptions, classifications and analyses.

Longer Context : Handles over 25,000 words (≈8K‑32K tokens) for extended conversations and document analysis.

Safety : Improved factuality and reduced hallucinations compared with previous models.

OpenAI also plans to release APIs for GPT‑3.5 Turbo, DALL·E 2 and Whisper, and to allow developers to fine‑tune GPT‑4 and GPT‑3.5 Turbo later this year.

Deprecated APIs

The company will retire several older Completion models, marking them as “legacy” in the documentation; they will become unavailable after January 4 2024.

Limitations

Despite advances, GPT‑4 still hallucinates, can produce biased outputs, and may make reasoning errors. OpenAI is working on customizable models to increase diversity and reduce bias.

Access to GPT‑4 is currently through ChatGPT Plus ($20/month) and via API for paid users, with higher pricing than GPT‑3.5. Microsoft’s Bing now runs on GPT‑4, and enterprises such as Mercedes‑Benz and Morgan Stanley are integrating it into production and wealth‑management workflows.