OpenAI resets ChatGPT and Codex limits for GPT-5.6 Sol

Whenever a frontier AI model drops, power users brace for the same inevitable buzzkill: burning through their weekly rate limits in twenty minutes flat.

OpenAI decided to bypass that launch-day anxiety entirely. To celebrate the debut of its flagship GPT-5.6 Sol, the company reset rate limits twice across both ChatGPT Work and Codex over a one-hour promotional window, giving users free rein to push complex agentic workloads without watching the meter.

The public release also closes a tense regulatory chapter. Following roughly two weeks of restricted preview access granted only to government-approved groups, OpenAI received approval from the Trump administration to roll the model out to the world. CEO Sam Altman called it the best model the lab has ever produced.

The desktop superapp and the pricing freeze

The GPT-5.6 lineup lands in three distinct flavors: Sol, Terra, and Luna. Flagship Sol finishes just a hair behind rival Fable on the AA Intelligence Index, but it takes first place in agentic coding while gaining notable jumps in computer use, visual design, and cybersecurity.

OpenAI also tucked an intriguing technical detail into the release: the smaller Luna tier was autonomously post-trained by Sol itself. For users demanding maximum reasoning, a dedicated Ultra mode was also introduced.

On the business side, OpenAI chose not to hike rates. Pricing holds flat against GPT-5.5, running from $5 and $30 per million tokens on Sol down to $1 and $6 on Luna. As Altman put it, every enterprise customer right now is obsessing over infrastructure spend.

The bigger software story is ChatGPT Work, OpenAI's direct answer to Anthropic's Claude Cowork. The tool wraps the power of the Codex engine into an accessible platform for everyday non-programmers, and it comes bundled into a revamped ChatGPT desktop app alongside a built-in browser and native computer controls.

Through a unified plugins directory, Work can plug straight into Slack, Gmail, Google Drive, calendars, and enterprise CRMs to churn out finalized presentations, spreadsheets, or lightweight web apps. Windows and Mac users worldwide, including those on free tiers, are gaining immediate access.

Why it matters

Lately, power users have spent half their time juggling token conservation setups just to keep Claude from locking them out mid-afternoon. By doubling down on rate resets and locking API prices at older generation rates, OpenAI is leaning heavily into Anthropic's biggest pain point: availability and cost.

For knowledge workers and operations teams, ChatGPT Work bridges the gap between raw code-generation models and functional desk work. Letting a system agent ingest files across local apps to produce full presentations without writing a single line of script is worth testing right now, especially on the newly merged desktop client.

We think merging Codex into the core ChatGPT desktop environment is a brilliant distribution play. Benchmark leads fade in weeks, but cementing a desktop footprint with native browser navigation and computer use gives OpenAI an operational moat that simple chat interfaces cannot match.

The real risk, however, lies in autonomy and security. Granting an AI agent broad visibility over internal mail, calendars, and CRM platforms sounds fantastic until an unintended execution loops across live business records. OpenAI has cleared the regulatory bar, but proving reliable execution across unconstrained desktop environments remains an uphill fight.