GPT-5.6 SOL: THE GOLD STANDARD FOR KNOWLEDGE WORK
TL;DR
Speed and steerability beat raw intelligence: The team found 5.6's quick iteration and responsiveness made it their default tool, even when Fable scores higher on benchmarks.
The app merge solves a real problem: OpenAI consolidated ChatGPT and Codex into one app because non-engineers at OpenAI were already using Codex for finance, recruiting, and sales work.
Writing quality has flipped: Katie Parrot now prefers OpenAI models for writing because they take feedback better, while Anthropic models over-fixate and become less usable during iteration.
Fable still wins on complex coding: On the senior engineer benchmark, Fable scored 91/100 versus 5.6's 56/100, but few tasks need that level of quality.
Design has a warm paper aesthetic: 5.6 produces consistent but predictable designs with warm paper tones, while Fable takes more creative risks.
Loop workflows are now accessible to non-developers: Knowledge workers can now delegate tasks and build systems rather than doing work directly, a workflow previously limited to programmers.
The Breakdown
The Merge: Two Apps Become One
The Every team gathered for a live vibe check of GPT-5.6 SOL, coinciding with OpenAI's merger of the ChatGPT and Codex desktop apps into a single application with two tabs: ChatGPT Work (code hidden) and ChatGPT Codex (code visible). Dan Shipper called it the gold standard for knowledge work, praising its combination of power, speed, and usability. The team had briefly lost access during what they called the 'government detour' and felt genuinely relieved to have it back.
Operations Work Gets Real
Ariel Shipper, head of operations, tested 5.6 on a real task: compiling 46 CSVs from an email into a unified analysis. Where 5.5 failed to find the email and asked redundant questions, 5.6 located the correct message, previewed the data, asked intelligent clarifying questions, and produced usable output on the first try. She noted that for multi-threaded knowledge work, she wants one tool that just works without thinking about which mode she's in.
Writing and the Feedback Loop
Katie Parrot, who wrote the written vibe check, drafted the review through 24 turns with 5.6 over six hours. She highlighted that while 5.6's first-pass prose isn't necessarily stronger than Anthropic's, it takes feedback dramatically better. She's noticed Anthropic models over-fixating and becoming less usable, while 5.6 stays focused on the larger task during iteration. She now spends more time in the Codex app than in her browser.
Coding: Fast Enough, Not Fable-Level
On the senior engineer benchmark (rewriting vibecoded slop), Fable scored 91/100 while 5.6 scored 56/100. The gap came from 5.6 not deleting or simplifying enough code. But the team emphasized that few tasks require Fable-level quality, and 5.6 serves as an excellent sub-agent for Fable at a fraction of the cost. Dan built a food tracker that scans his photos and voice memos automatically, something previously impossible without an Apple Photos API.
Design: The Warm Paper Problem
Mike Taylor and Dan debated design quality. Mike preferred 5.6's warm paper aesthetic for his eval packs (HTML pages that explain what the AI did). Dan showed a head-to-head comparison where Fable produced a more sophisticated diagram from the same prompt. The team agreed 5.6 has improved design but tends toward predictable warm paper tones, while Fable takes more creative risks.
OpenAI Guests on the Merge
OpenAI's Dom and Roman joined to discuss the merge. They revealed that every team at OpenAI, from finance to recruiting, had been using Codex daily, which motivated consolidating the apps. They demoed a new visualizations feature where 5.6 inspected video frames and rebuilt interactive demos. They emphasized that computer use is now 'basically solved' with the Chrome extension.
Finance at OpenAI: A Power User Demo
Kyle Cobber from OpenAI finance showed a custom dashboard he built over a month that compresses five days of monthly compute accounting into five hours. The system pulls from data lakes, Excel, and Google Sheets, then outputs to a hosted app with virtual slides. His advice: don't try to one-shot complex workflows. Instead, build skills incrementally as you work through the process, treating Codex like an analyst you're training.
Claire Vo's Blind Taste Test
Claire Vo ran blind taste tests comparing 5.6 and Fable on PRD writing and design. She found 5.6 took more creative risks in design while Fable produced denser, more technical outputs. Her key distinction: Fable optimizes the model-to-model interface (agent orchestration), while 5.6 optimizes the human-to-model interface (collaboration). She values effectiveness over raw intelligence and called 5.6 a 'power teammate.'
Was This Useful?
Share
Keep Reading
Make Alcreon Yours
Tune your feedFive quick questions, and the feed ranks what matters to you first.Or just get notified
The weekly Echo. Signal worth keeping in your inbox.
Every new piece, announced on X.
Read Next
See all
Playbook
The Retirement Email Isn't a Warning
Model retirements now arrive every few weeks; the config-eval-rehearsal loop turns each deprecation email from a fire drill into an afternoon swap.
Playbook
The Cheapest Model That Passes
OpenRouter lists 400 models behind one API. The fix for choosing isn't a better leaderboard, it's a four-step protocol that ends in a real eval.
Playbook
Cheap Models, Hard Tasks
Most agent workflows route every step to the frontier model by default. The bill scales with how chatty the agent gets, even when most steps don't need that brain.