GPT-4o

GPT-4o is OpenAI’s natively multimodal model, released on May 13, 2024, that processes text, vision, and audio within a single neural network, eliminating the separate text-only and vision pipelines used by predecessors like GPT-4 and GPT-4V. In a snapshot tracked on February 16, 2026, it achieved an MMLU-Pro score of 73.0, an Arena Elo of 1286, and a SWE-bench Verified score of 38.4, with pricing fixed at $2.50 per million input tokens and $10.00 per million output tokens. By May 2024, it served as the default model for the ChatGPT free tier globally, delivering average response latencies of 320 ms for audio and 230 ms for text—faster than GPT-4 Turbo’s typical 500–800 ms text latency, at roughly half the per-token cost. GPT-4o matters because its February 2026 benchmark snapshot establishes a precise, dated reference point for evaluating how OpenAI’s flagship multimodal model balances cost and performance over time, and its unified architecture underpins real-time voice and vision applications requiring sub-second, cross-modal responsiveness.

Two-hop subgraph: this entity, every entity it directly relates to, and every entity those neighbors relate to. Drag a node, scroll to zoom, click to inspect — or click any neighbor and re-center the atlas there.