GPT 5.6: A Major Leap in AI Model Performance and Ecosystem Developments

OpenAI released GPT 5.6, described as a "monster leap" over GPT 5.5, almost akin to the jump from GPT 3.5 to GPT 4. The model delivers near-fable-level intelligence at significantly lower costs. GPT 5.6 soul pro generates outputs costing $0.77 for 45,000 tokens in about one minute, compared to GPT 5.5 pro at $4 and six minutes for similar tasks. Benchmarks show GPT 5.6 outperforming most competitors except Claude fable by a narrow margin in GDP Val, excelling at browser-based tasks, and tying or surpassing mythos in GPQA. On beauty bench, GPT 5.6 soul pro secured the top spot, with its variants occupying the next three.

OpenAI also launched "chat GPT work," a super-app merging Codecs, Chat GPT, and Atlas browser into a unified Chat GPT app. Work mode serves as a personal assistant, connecting to Gmail, Google Drive, Slack, Obsidian vaults, and more. The app can visualize second brain structures and build tailored control towers for task management, offering recurring scheduling and cross-platform accessibility.

New features include the ability to prompt site creation directly, automated game builds (e.g., soul bonk clone done in a single prompt and hosted instantly), and responsive UI adjustments between work and codecs modes. The app sidebar reflects coding and productivity options based on setting.

X AI released Grok 4.5, which ranks closely with current leaders on deep sweet (Grok at 62%, Fable at 66%, GPT 5.6 at 72.7%) and terminal bench (Grok at 83, GPT 5.6 at 88, soul ultra at 91.9%). Grok can be installed locally via CLI and offers slide deck generation and visual web demos, with API pricing at $2 per million input tokens and $6 output tokens.

Meta introduced Muse Spark 1.1, matching previous generation models (terminal bench at 80, deep sweet at 53.3), with Muse Image stirring controversy for its Instagram-based personal image generation. Pricing aligns with Grok and GPT 5.6 Terra models. Muse Spark's image advancements show significant improvement compared to older Meta models.

Anthropic extended Fable access, reset weekly limits, and launched Claude co-work for mobile/web cloud execution. Claude now offers a "reflect" feature (usage analytics) and published research on "J space"—AI subconscious thought tracking. Google released video remix with Gemini Omni for stylized clips, and ByteDance launched Seed Dream 5.0 pro for layered, photorealistic, and infographic image generation.

The speaker emphasizes personal productivity gains and website development using both Fable and GPT 5.6, noting GPT 5.6's superior bug discovery. Fable is seen as a "thinking power" tool for tough architectural decisions, while GPT 5.6 is ideal for bulk execution. The week marks "the biggest jumps" in model capability since GPT 4, exciting developers for rapid platform building and automation.

Rapid Fire Additions

• Fable access extended from July 7th to July 12th, with weekly limits reset.
• Claude co-work now runs tasks in the cloud, accessible across devices.
• Claude "reflect" tracks peak activity; new research into "J space" (AI subconscious).
• Google video remix released for AI-powered clips using Gemini Omni.
• ByteDance Seed Dream 5.0 pro image model excels in infographics and interactive edits, now on Leonardo.

Comparative Notes

Fable is best for smart architecture, GPT 5.6 excels at bulk assignments, and the combination ensures bug detection and security. The speaker rebuilt the Future Tools website primarily with Fable and GPT 5.6, highlighting how closely each met intended goals with minimal guidance.