"GPT-5.6 is here but also the long awayited Superapp the tl;dr The model side is impressive, but expected: GPT-5.6 Sol is strong across coding, browsing, cyber, science, long-context and agentic work. On several evals, OpenAI claims it either sets new SOTA or comes very close while..." — Chubby
...using fewer tokens, less time, or lower estimated cost. However, even outperforms Fable 5 on several Benchmarks, obviously not on all tho (see below). Some notable numbers: • Agents’ Last Exam: GPT-5.6 Sol reaches 52.7%, ahead of GPT-5.5, Claude Fable 5 and Opus 4.8 • Terminal-Bench 2.1: Sol Ultra reaches 91.9%, above Claude Mythos 5’s reported 88.0% • BrowseComp: Sol Ultra reaches 92.2% • OSWorld 2.0: Sol reaches 62.6%, ahead of Opus 4.8 • Artificial Analysis Coding Agent Index: Sol scores 80, ahead of Fable 5 • SEC-Bench Pro: Sol Ultra reaches 74.3% But the app layer is what makes it interesting: ChatGPT Work can pull context from docs, Slack, Notion, Microsoft 365 and Google Drive, then turn that messy context into actual outputs: decks, documents, spreadsheets, dashboards, visualizations and interactive explanations. https:// chatgpt.com/c/6a4fd9b3-0a8 0-83eb-bae7-b58211c59cc5 … — Chubby
Source: https://x.com/kimmonismus/status/2075271465964798147