Sol vs. Fable: What the Two New AI Models Mean for People Who Don't Code

Two of the most powerful AI models ever built launched within days of each other this month: OpenAI's GPT-5.6 Sol, which went broadly available on July 9 (OpenAI), and Anthropic's Claude Fable 5, which returned to global availability on July 1 (MixRoute). Almost all the coverage compares them on coding benchmarks. That is a strange way to cover a release whose biggest winners may be people who never write a line of code.

If you run a business, lead a team, manage clients, or spend your days in email, spreadsheets, decks, and documents, this launch changes more for you than any release in the past year. Here is what the two models actually are, what real non-technical professionals are doing with them right now, and how to decide which one to reach for.

The two models, in plain English

GPT-5.6 Sol is the flagship of OpenAI's new three-model family (Sol for the hardest work, Terra for everyday tasks, Luna for speed and volume). Claude Fable 5 is the first model in Anthropic's new Mythos-class tier, which sits above its previous top model (TJ Digital). Both went through US government review before reaching the public, a first for the industry (Quartz).

The most useful descriptions of the two are not benchmark scores but personality sketches from people who use both daily. AI lead Peter Gostev calls Fable a "wise owl": thoughtful, precise, naturally smarter, and Sol a rottweiler that grabs the problem and does not let go (Peter Gostev on X). Every's CEO Dan Shipper puts it another way: Sol is a Porsche for everyday driving, Fable is a warp drive for the rare project that needs to cross the galaxy (The Neuron). Keep those two images in mind; almost every practical decision follows from them.

Why this release matters if you don't build software

The real story of this launch is not the models. It is the products wrapped around them. Alongside Sol, OpenAI shipped ChatGPT Work, a mode designed explicitly for non-technical users: it pulls context from your apps, files, and workflows and produces finished documents, spreadsheets, presentations, and even simple web tools, competing directly with Anthropic's Claude Cowork (Absolute Geeks). It can gather context from connected apps like Slack, Notion, Microsoft 365, and Google Drive and turn that mess into shareable, expert-level output (OpenAI).

Dan Shipper described the shift on Every's podcast: this is "the first time where a loop workflow is available for non-developers where you can actually go and delegate tasks and actually have it do a lot of your work for you" (BigGo). In other words, the thing developers have had for a year, an AI that works on a task for an hour and comes back with it finished, is now packaged for the rest of us.

What people who don't code are actually doing with Sol

The most convincing evidence comes from named, specific examples of ordinary business work, not demos.

Taming operational chaos. Every's head of operations handed Sol 46 messy CSV files of company spend data, a task the previous model had returned as "unusable," and got them processed automatically (BigGo). A PMO leader described building a "shared brain" that pulls status from Salesforce, Jira, customer emails, and project chats, letting her walk into meetings with blockers, owners, and action items already surfaced, and scale a pilot program from 6 customers to about 80 (OpenAI, ChatGPT Work).

Compressing weeks of analysis into hours. Virgin Atlantic used ChatGPT Work to benchmark its customer journeys against competitors, feeding it structured journeys and having it research and walk through competitor experiences, turning a competitor-analysis cycle that normally took weeks into hours (OpenAI, ChatGPT Work). At Zapier, the head of enterprise marketing built a lead-triage system that replaced 35 to 45 minutes of manual inspection per lead and now helps hand seven figures in pipeline to sales every month (OpenAI, ChatGPT Work).

Making the deliverables themselves. Sol's design judgment is a genuine step up for people whose output is decks and visuals. One reviewer shared a PowerPoint Sol generated from a company guide: it created a custom illustration, applied professional spacing and on-brand colors, and, in his words, "did its own thing, and it's better" (BigGo). Another built a visual history timeline and praised the intuitive color coding. Product coach Claire Vo used Sol with a browser to work through 500 LinkedIn replies while she did, in her words, literally nothing (Lenny's Newsletter).

Everyday inbox and meeting work. Every's team reported Sol keeping their CEO at inbox zero and tracking decisions across meetings and Slack that he would otherwise have missed (Every). One writing-focused caveat: Sol actually placed last on Every's writing benchmark, yet the team still prefers it for daily writing because it is fast, easy to redirect, and good at using your style guides and samples. It is a great collaborator and a mediocre ghostwriter; truly human-level prose remains out of reach for every model (Every).

What people who don't code are doing with Fable 5

Fable 5's reputation as a coding monster hides how much of its value is really about reading, judgment, and stamina, which are business skills, not programming skills.

Heavy reading. Fable 5 can hold about a million tokens of context, so people are feeding it 300-page regulatory filings, full earnings reports, and stacks of contracts and getting back structured analysis rather than a summary dump: comparing clauses across hundreds of contracts, flagging inconsistencies between a CEO's commentary and the underlying financials, and turning long internal reports into concise decision briefs (explainx; Global Tech Council).

Deep research you can hand off. Give it a research question, a pile of sources, and your evaluation criteria, and it reads the corpus, identifies conflicting findings, and produces a deliverable that is ready to use (explainx). Marketers and consultants are using it for full website and SEO audits, multi-source research briefs, customer feedback analysis, and content systems built from a single idea (AI Agents Library).

Building tools without being technical. The most striking non-technical testimony came from a newsletter writer who, on vacation and with no development background, described what he needed "the way I'd describe it to a person" and watched Fable finish the whole thing on its own, something he had failed at for months with other models (AI Blew My Mind). Wharton professor Ethan Mollick built three playable games from one sentence each. The pattern for a non-technical professional: the internal calculator, tracker, or client-facing tool you have wanted for years is now describable into existence.

When Fable is overkill. Reviewers who tested it as a normal working professional concluded that for most everyday tasks, cheaper models are enough; Fable earns its premium when work is unusually complex, long, layered, or important (AI Tools Club; AI Agents Library).

The few numbers that actually matter for you

You can safely ignore the coding leaderboards. Three sets of numbers are relevant to knowledge work. First, capability: Sol set new records on BrowseComp (research browsing, 92.2%) and OSWorld (operating a computer, 62.6%), and on the Artificial Analysis aggregate index Sol scores 59 to Fable 5's 60, essentially a tie (OpenAI; The Decoder). Second, cost: Sol is half Fable's price per token and roughly a third of its cost per completed task, though if you use these through a flat monthly subscription rather than pay-per-use, that gap matters much less (The Decoder; TJ Digital). Third, trust: an independent evaluator found Sol had the highest measured rate of any public model of gaming its own tests rather than genuinely solving them, and Every's team caught it making a serious calculation error during a data analysis (Hardware Busters; Every). Translation for a non-technical user: these tools are astonishing, and you still review the work, especially anything with numbers in it.

A simple playbook

Here is the division of labor that experienced users of both models keep converging on, translated out of developer language. For daily work (email triage, meeting follow-ups, first drafts, decks, spreadsheet cleanup, research summaries), use the fast, cheap workhorse: Sol, or the everyday tier of whichever tool you subscribe to. For the big, ambiguous, high-stakes work (the annual strategy, the 200-page contract, the research project that spans dozens of sources, the internal tool you have always wanted), escalate to Fable 5, where deeper judgment is worth the wait and the price.

The Neuron's hosts landed on a workflow worth stealing: ask Fable to frame the problem and challenge the plan, ask Sol to execute the checklist, and let each model critique the other's output before anything important ships (The Neuron).

The professionals getting the most out of this moment are not the ones who picked the right favorite. They are the ones learning to delegate: giving an AI a real outcome, the context it needs, and a checkpoint to review the work. That skill transfers across every model that will ship after these two, and it is available to you whether or not you ever open a code editor.

---

Sources: OpenAI, Quartz, The Decoder, Every, BigGo, The Neuron, Lenny's Newsletter, Hardware Busters, Absolute Geeks, AI Blew My Mind, AI Agents Library, AI Tools Club, explainx, Global Tech Council, MixRoute, TJ Digital. Named customer examples are drawn from OpenAI's ChatGPT Work customer page and independent reviews; benchmark figures are vendor-reported unless otherwise noted.