ChatGPT Work with GPT-5.6: What Heavy AI Users Need to Know About Access and Costs
OpenAI just launched ChatGPT Work with GPT-5.6 Sol, Terra and Luna. Here is the real cost math for heavy AI users and how it stacks up against Claude Cowork.
On July 10, 2026, OpenAI rolled out two things at once: a public release of GPT-5.6 after its brief government-only preview period, and a new product called ChatGPT Work. If you spend several hundred dollars a month on AI and you are watching both the OpenAI and Anthropic ecosystems, this is the one story from the past 48 hours that directly affects your toolchain and your bill.
Here is what you need to know.
What ChatGPT Work Actually Is
ChatGPT Work is not a new standalone app. It is a unified agent interface that combines the conversational layer of ChatGPT with the coding and automation capabilities of Codex, all running on the GPT-5.6 model suite. OpenAI describes it as allowing “the everyday non-technical user to take advantage of Codex’s capabilities for non-coding tasks.” In practice, it means Codex-level task execution is now surfaced in the same chat window where you already write prompts.
It connects to Slack, Gmail, Google Drive, calendars, and CRMs via a “unified plugins directory.” It gathers context from apps and files you authorize, then produces finished artifacts. OpenAI says it can “turn long email threads into action items, research docs into polished deliverables, and sprawling project notes into structured roadmaps.”
The direct competitive frame is explicit: OpenAI built ChatGPT Work as a counter to Anthropic’s Claude Cowork, which similarly combines Claude and Claude Code in a single agent interface.
The GPT-5.6 Model Tiers: Sol, Terra, and Luna
ChatGPT Work runs on the full GPT-5.6 suite, which has three public tiers:
- Sol ($5/M input, $30/M output): The flagship. OpenAI is positioning Sol to “set a new standard for intelligence and efficiency” in coding, cybersecurity, and science. Computer use capabilities are included. Sol is the tier that received the Trump administration’s greenlight for public rollout after a restricted government-only preview that raised some METR evaluation flags.
- Terra ($2.50/M input, $15/M output): The mid-tier. Designed for most workplace tasks where Sol-level power is not necessary.
- Luna ($1/M input, $6/M output): The entry tier, positioned as a lower-cost alternative to competitors. OpenAI is specifically marketing this against the backdrop of “industry-wide AI money squeeze and AI lab costs being passed onto customers.”
For context, that is Claude Sonnet 5 intro pricing at $2/$10 per million tokens. Luna undercuts it. Sol is priced in the same range as Claude Opus 4.8 ($15/$75). Terra sits in between.
Who Gets Access and When
This is the part that matters most for heavy users operating across multiple subscription tiers:
- Desktop app (Mac and Windows): All users including free tier get immediate access to ChatGPT Work and GPT-5.6.
- Mobile and web: Pro, Enterprise, and Edu users got access first. Plus and Business users receive access “over the next few days.”
- Codex remains a separate product: OpenAI clarified that the standalone Codex app is “here to stay” despite the ChatGPT Work launch. Work integrates Codex capabilities but does not replace the dedicated Codex Cloud environment for teams running agentic pipelines at scale.
If you are on a Max plan at $200/month, you do not automatically get ChatGPT Work. OpenAI’s plan hierarchy has not aligned ChatGPT Work access with the Max tier in the way heavy Claude users might expect. The desktop app bypass is your fastest path to access if you are not on Pro or Enterprise.
The Cost Math for Heavy Users
Let’s run three scenarios for a user spending around $400/month on AI:
Scenario 1: You are already on ChatGPT Pro ($200/month) GPT-5.6 Terra and Luna are included in your usage window. Sol likely has per-session limits at Pro tier, similar to how GPT-5.5 Ultra was handled. You get ChatGPT Work access on mobile and web immediately. Your cost stays flat unless Sol burns through your quota faster due to the computer use overhead.
Scenario 2: You are on ChatGPT Plus ($20/month) for ChatGPT plus Claude Max ($200/month) Plus users get Work access “over the next few days” per OpenAI. But the real question is model tier. At Plus pricing, you are getting Luna and Terra in Work, not Sol. If you need Sol-level performance for coding tasks, you are either upgrading to Pro or paying Sol API rates directly ($5/M in, $30/M out), which is identical to the GPT-5.6 Sol pricing released two days ago.
Scenario 3: API-first heavy user Luna at $1/$6 is the most interesting new option. For tasks where GPT-5.5 or Claude Sonnet 5 were handling mid-tier reasoning, Luna is price-competitive. For high-stakes autonomous coding runs, Sol at $5/$30 is expensive relative to Claude Sonnet 5 intro pricing, but on par with Claude Opus 4.8 fast mode. The 54% token efficiency improvement Altman cited for agentic coding (announced with GPT-5.6 two days ago) matters here: if Sol genuinely uses fewer tokens per equivalent output, the effective cost per task may be lower than the per-token rate suggests.
OpenAI’s Internal Turbulence: Why It Matters for Reliability
Heavy AI users care about reliability almost as much as pricing. In the same week that ChatGPT Work launched, OpenAI saw three notable departures: Joshua Achiam (chief futurist, nine years), Fidji Simo (AGI head, stepping down due to illness), and Johannes Heidecke (safety oversight lead). Greg Brockman is consolidating product authority ahead of the company’s prospective IPO.
These departures do not directly affect model performance or API uptime. But they are context for understanding where OpenAI’s internal attention is focused. The Sol tier’s METR evaluation flags, combined with the safety team contraction, are worth tracking if you are running Sol for anything beyond standard productivity tasks. The “government greenlight” framing for Sol’s public rollout means OpenAI moved forward despite those flags rather than waiting for resolution.
If you rely on OpenAI’s API for production workloads, keep an eye on Sol’s reliability metrics over the next few weeks as it scales to general availability.
ChatGPT Work vs Claude Cowork: The Real Comparison
Both products are betting on the same workflow premise: a single interface that handles chat, code generation, and multi-step task execution across connected work apps.
| Feature | ChatGPT Work | Claude Cowork |
|---|---|---|
| Models | GPT-5.6 Sol / Terra / Luna | Claude Sonnet 5, Claude Opus 4.8, Fable 5 |
| Integrations | Slack, Gmail, Drive, Calendar, CRMs | Asana, Notion, Slack, browser-native tools |
| Coding engine | Codex (integrated) | Claude Code (integrated) |
| Desktop access | All users immediately | Mobile added July 7 |
| Entry API cost | Luna $1/$6 per MTok | Sonnet 5 $2/$10 per MTok (intro) |
| Flagship API cost | Sol $5/$30 per MTok | Opus 4.8 $15/$75 per MTok |
On raw API pricing, OpenAI’s Sol is cheaper than Claude Opus 4.8 for flagship-tier work, and Luna is cheaper than Claude Sonnet 5 at intro rates. The question is whether the model quality justifies the switch for your specific workloads. The 54% token efficiency claim from Altman is the variable that could change the math significantly, but it needs to be verified in production.
What to Do This Week
If you are evaluating model costs for agentic pipelines: Run a benchmark of GPT-5.6 Luna against Claude Sonnet 5 on your actual task types before the intro Sonnet pricing expires August 31. Luna vs Sonnet 5 is the most interesting cost comparison in the mid-tier right now.
If you are on ChatGPT Pro or Enterprise: ChatGPT Work is live for you now on web and mobile. It replaces the separate Codex Cloud interface for most non-coding autonomous tasks. Map your current Codex Cloud usage to see if Work covers it or if you still need both.
If you are a Plus subscriber: Wait for the Work rollout to your tier, then test the desktop app first since it gives immediate access regardless of web/mobile timing. Track which model tier you actually get in Work at Plus pricing.
If you are managing a multi-provider AI budget over $300/month: Sol at $5/$30 is entering the range where a serious cost comparison with Claude Opus 4.8 makes sense. The efficiency claims need field validation, but the pricing alone suggests Sol is worth testing for coding and research tasks where you currently default to Opus.
The ChatGPT Work launch is OpenAI’s most direct attack yet on Anthropic’s Cowork product, and the GPT-5.6 pricing structure gives heavy users more granular options than the previous generation. The turbulence in OpenAI’s leadership is a factor to monitor but not a reason to change your toolchain today.
Now available
Stop guessing your AI limits
The Mac app and web dashboard watch your Claude, ChatGPT, Gemini and more, and warn you before quotas hit.