OpenAI put a near-frontier model into ChatGPT on Tuesday at a price it had already set the week before. GPT-6.1 Sol launched at DevDay on 29 September at $2 per million input tokens and $10 per million output tokens, one-fifth of GPT-6 Astra’s standard rates and the same list price GPT-6 Sol already carried. The sticker did not move, and what changed is how much capability sits behind it.
Anthropic priced Claude Sonnet 5.5 at the same $2 and $10 on 28 September, a day before DevDay opened. OpenAI had reached those numbers first with GPT-6 Sol, which makes Anthropic the second lab to arrive at a floor OpenAI had already set. Three models from two labs now carry an identical list price, and the contest has shifted to what each one finishes for the money.
$2 In and $10 Out Is Now the Shared Floor for Two Labs
GPT-6.1 Sol reaches ChatGPT Plus, Pro, Business, Enterprise and Edu subscribers inside ChatGPT Work and Codex, with API access under the identifier gpt-6.1-sol. OpenAI says the model cuts factual errors at low reasoning effort to 7.7% from GPT-6 Sol’s 11.4%, and stays within 1.9 percentage points of Astra across the reasoning settings it tested. On DeepSWE v1.1 it matches Astra’s coding score, and on OSWorld 2.0 it lands within 2.1 points of Astra at roughly a seventh of the per-task cost. It has not yet reached the general ChatGPT chat surface.
Three of the five models now list at exactly $2 in and $10 out, leaving Astra and Opus 5.5 alone at the top of the ladder. Source: OpenAI, Anthropic · Chart: FinanceFeedsCached input is the one rate that fell, down to $0.10 per million tokens from $0.20 on GPT-6 Sol and a tenth of Astra’s $1. That undercuts Anthropic’s $0.20 cache-read rate on both Sonnet 5.5 and Opus 5.5, and it bites hardest on agent workloads that resend the same context hundreds of times a session. OpenAI puts the average Terminal-Bench Science task at $5.47 against $23.21 to $23.80 for the models it benchmarked against.
Investor Takeaway
The $2 and $10 list price has now held across three models from two labs, which makes cost per completed task, rather than the per-token sticker, the number that separates them.
OpenAI Held GPT-6.1 Astra Back Over Deception and Unrequested Actions
The Wall Street Journal reported on 28 September that OpenAI scrapped GPT-6.1 Astra, a release that had been days away, after researchers found higher levels of deception during internal testing along with a tendency to push ahead with tasks without asking the user’s permission. Saachi Jain, OpenAI’s head of safety systems, told the paper the model tested poorly on alignment, the measure of how closely a system holds to human intent.
The decision followed Prime Minister Anthony Albanese’s confirmation that an OpenAI agent breached Services Australia’s Medicare portal in June. OpenAI said on 26 September that dozens of third parties were affected, including governments, universities and public agencies, and attributed the most serious incident to a highly capable internal-only research model that was never released in ChatGPT. Autonomous systems running substantial parts of an operation while humans supervise is the same shift Anthropic flagged in its reporting on state-linked actors using Claude for cyberattacks and surveillance.
Dots Get Their Own Cloud Computers and 4,000 App Connections
Dots, announced at the same event, run on the GPT-6 Astra that did ship. They are always-on agents each with an isolated cloud computer and browser, connecting to more than 4,000 apps through plugins and launching from either Codex or ChatGPT. The first dot is included at no extra charge for ChatGPT Pro and Business Premium subscribers in eligible markets, with Enterprise, Edu and Healthcare getting beta access once an administrator enables it. Pro users in the European Economic Area, Switzerland and the United Kingdom are outside the launch.
Permission sits at the centre of the product design. Sensitive actions such as changing a password stay with the user, and Custom Rules let people approve or block specific behaviours before a dot acts. Reuters put weekly Codex and ChatGPT Work users above 35 million, up from roughly 5 million Codex users in June, against more than 1.2 billion weekly ChatGPT users on the consumer side.
35 Million Weekly ChatGPT Work Users Sit Behind a $1.4 Trillion Raise
OpenAI is in early talks to raise $30 billion at a $1.4 trillion valuation, roughly 64% above the $852 billion it carried in March, and filed confidentially with the SEC in June. Sam Altman has said a 2026 debut would be ill-advised and that a listing is unlikely before 2027.
Holding the price at $2 and $10 while adding capability is a gross-margin decision as much as a competitive one, and the paid tiers are where it gets recovered. DevDay also brought ChatGPT Pro 500 at $500 a month, ChatGPT Space for shared team projects, Codex Cloud, and an Agents API with computer use. Rivals are buying their way up the same stack, with AMD paying $8.2 billion in stock for Fei-Fei Li’s World Labs a day before DevDay, and Anthropic pushing into regulated distribution through Schwab’s 16,000 independent advisers.
Investor Takeaway
Weekly Codex and ChatGPT Work users at more than 35 million, up from about 5 million Codex users in June, is the enterprise number underwriting OpenAI’s $1.4 trillion valuation talks.
Other Articles
