GPT-5.6 Sol beat Claude Fable 5 by 13.1 points on Agents' Last Exam, and that number shows why OpenAI's latest launch is aimed less at model bragging rights than enterprise displacement.
XOOMAR Intelligence
Analyst Take
The new GPT-5.6 lineup, paired with ChatGPT Work, looks like OpenAI's attempt to box in Anthropic on three fronts at once: quality, speed, and workplace distribution, according to ZDNet. XOOMAR analysis: the sharper move may not be the model upgrade itself. It's the packaging of agentic work into a product that can sit inside daily enterprise workflows.
GPT-5.6 turns OpenAI's model race into an enterprise software ambush
OpenAI introduced three GPT-5.6 models: Sol, Terra, and Luna. ZDNet describes Sol as the flagship model aimed at Anthropic's Fable 5, Terra as the mainstream default model, and Luna as the lighter instant-style model.
That tiering matters because OpenAI is no longer selling one abstract idea of "better AI." It's selling a menu. Buyers can match cost, speed, and capability to the task instead of defaulting to the biggest model for everything.
| Model | Role described in source | Comparable GPT-5.5 tier |
|---|---|---|
| GPT-5.6 Sol | Flagship model aimed at Fable 5 | Thinking mode |
| GPT-5.6 Terra | Mainstream default model | GPT-5.5 |
| GPT-5.6 Luna | Lighter, less capable model | Instant model |
ChatGPT Work is the real ambush. ZDNet says it lets AI access a user's desktop and browser to perform tasks, a capability previously embedded in Codex, OpenAI's agentic programming tool. Separating that into a workplace assistant makes it legible to sales, finance, marketing, operations, and executive teams.
XOOMAR analysis: OpenAI is moving from selling intelligence to selling productivity infrastructure. If ChatGPT Work becomes the place where employees build decks, reconcile spreadsheets, organize notes, and chase CRM follow-ups, Anthropic has to compete not only on reasoning quality but on daily habit.
For related XOOMAR coverage on the product angle, see ChatGPT Work Takes the Wheel on Hours-Long Office Tasks.
The price and speed math behind OpenAI's challenge to Anthropic Claude
OpenAI's benchmark claims are unusually pointed. ZDNet reports that GPT-5.6 Sol scored 53.6 on Agents' Last Exam, beating Claude Fable 5 by 13.1 points. The benchmark tests long-running, multi-step professional workflows across 55 fields.
OpenAI also says GPT-5.6 Sol came within one point of Fable 5 on the Artificial Analysis Intelligence Index while completing tasks in 61% less time. On the Artificial Analysis Coding Agent Index, Sol reached 80, or 2.8 points above Fable 5.
The cost claim is just as aggressive. At medium reasoning, OpenAI says Sol beat Fable 5 by 11.4 points at roughly one-quarter the estimated cost.
"GPT‑5.6 was the strongest model we evaluated on our agentic code-review tests. On our apples-to-apples internal and external PR benchmarks, it beat GPT‑5.5 on F1 while using roughly 3x fewer tokens per PR and delivering about 2x lower median latency."
That quote comes from Itamar Friedman, Qodo's co-founder and CEO, cited in ZDNet. It points to the procurement question enterprises will actually ask: not "Which model wins the prettiest demo?" but "Which model finishes more work per dollar, with less waiting?"
The missing data still matters. ZDNet's source material does not provide full official details on cost per token, rate limits, uptime commitments, enterprise seat pricing, or context window size for GPT-5.6 and ChatGPT Work. Until those are clear, any hard buyer comparison remains incomplete.
ChatGPT Work targets the real enterprise bottleneck: daily AI use
ChatGPT Work can create spreadsheets, PowerPoints, and full websites, according to ZDNet. It can also work on complex projects for hours by breaking them into smaller tasks.
OpenAI's own positioning is explicit: give it work employees already understand. The company says users can ask it to "analyze a month-end budget variance, turn source materials into a marketing campaign brief, or prepare for a sales meeting."
The connectors are the strategic clue. ZDNet says ChatGPT Work works with Slack, Microsoft Teams, Google Drive, SharePoint, email, calendars, CRMs, and project trackers. OpenAI's ChatGPT Work page adds examples from teams at Virgin Atlantic, Zapier, NVIDIA, Shopify, and RingCentral.
One Zapier example is especially concrete. Angela Ferrante, head of enterprise marketing at Zapier, said ChatGPT Work traced customer touchpoints across Zapier's CRM, email, and other tools, found where follow-ups broke down, and generated a weekly executive dashboard that "revealed seven figures in potential sales."
XOOMAR analysis: that is the pitch CIOs understand. Not chatbot novelty. Repeatable workflows. Visible outputs. Less manual triage. But the trust issue is real. ZDNet's David Gewirtz described an OpenAI demo where ChatGPT Work reorganized Apple Notes as powerful but unsettling, writing that he has "trust issues at this level of delegation."
For a related OpenAI product shift, see XOOMAR's ChatGPT Voice Mode Stops Interrupting With GPT-Live-1.
From GPT-4 to GPT-5.6, OpenAI is packaging the platform
OpenAI's move follows a clear arc: consumer chat, developer tools, coding agents, and now workplace agents. ZDNet says OpenAI previously had separate ChatGPT and Codex desktop apps, and is combining the two into one app.
That matters because distribution often beats isolated technical superiority. If an AI assistant is already in the desktop app, connected to documents, chats, calendars, and source materials, model choice fades into the background for many users.
Anthropic remains the obvious target. ZDNet frames ChatGPT Work as OpenAI's answer to Claude Cowork, and says OpenAI's model announcement directly targets Anthropic's testing results and pricing levels.
Still, OpenAI has not removed every product question. ZDNet says it was not clear from the announcement whether ChatGPT Work will be folded into the newly combined ChatGPT and Codex app.
CIOs, developers, employees, and investors will grade GPT-5.6 differently
CIOs will focus on governance, access controls, monitoring, risk calibration, and predictable cost. ZDNet reports that OpenAI says GPT-5.6 includes safeguards combining model protections, real-time checks, monitoring, and access calibrated to trust and risk.
Developers will care about coding performance, latency, token use, and reliability in real workflows. The Qodo quote gives OpenAI a useful proof point, but broader API behavior still needs hands-on validation.
Employees will judge ChatGPT Work more simply. Does it save time? Does it make work easier to explain? Does it avoid creating a cleanup job after the automation runs?
Investors will read the launch as a monetization signal. XOOMAR analysis: OpenAI is trying to defend its lead by moving deeper into paid professional workflows, where usage can become habitual and harder to displace.
GPT-5.6 could trigger a price fight, but workflow lock-in is the bigger prize
The immediate buyer benefit is clear enough: if OpenAI's claims hold up in production, enterprises get stronger agentic performance at lower estimated cost and lower latency. That combination can change usage. Faster tools get used more often.
The risk is lock-in. Once one AI workspace captures files, prompts, project context, dashboards, notes, connectors, and approval patterns, switching providers becomes harder than swapping a model endpoint.
ChatGPT Work is rolling out on web and mobile for Pro, Enterprise, and Edu plans, with Plus and Business following over the next few days, according to ZDNet. OpenAI's own ChatGPT Work page says it is available to all plans on desktop today, with web and mobile rolling out over the next few days.
The next evidence to watch is practical, not theatrical: real enterprise pricing, admin controls, latency under load, connector reliability, and whether employees trust ChatGPT Work enough to hand it messy, high-value tasks. The winner won't be the lab with the cleanest benchmark chart. It will be the company that makes AI cheap enough, fast enough, and reliable enough that teams stop debating whether to use it.
The Bottom Line
- OpenAI is positioning GPT-5.6 as an enterprise replacement threat to Anthropic by competing on quality, speed, and productivity.
- ChatGPT Work could shift AI adoption from standalone chatbots into everyday workplace workflows.
- The tiered model lineup gives business buyers more flexibility to balance cost, speed, and capability by task.
GPT-5.6 lineup positioning
| Model | Role described in source | Comparable GPT-5.5 tier |
|---|---|---|
| GPT-5.6 Sol | Flagship model aimed at Anthropic's Claude Fable 5 | Thinking mode |
| GPT-5.6 Terra | Mainstream default model | GPT-5.5 |
| GPT-5.6 Luna | Lighter, less capable instant-style model | Instant model |
GPT-5.6 Sol performance lead over Claude Fable 5
Primary Sources & Disclosures
Written by
XOOMAR Insights Team
Research and Editorial Desk
The XOOMAR Insights Team pairs automated research with human editorial judgment. We track hundreds of sources across technology, fintech, trading, SaaS, and cybersecurity, cross-check the facts, and explain what happened, why it matters, and what to watch next. We do not just rewrite headlines. Every article is fact-checked and scored for reliability before it goes live, and we link back to the original sources so you can verify anything yourself.
