OpenAI has unveiled pricing for its latest large language model, GPT-6 Sol, setting the standard short-context rate at $2 per million input tokens and $10 per million output tokens. The new tier sits 80% below the company's GPT-6 Astra model on both input and output metrics, according to OpenAI's current model page. The development, confirmed on September 22, 2026, positions Sol as a mid-range option between the premium Astra and the ultra-low-cost Luna.
For Microsoft Corporation (NASDAQ:MSFT), which maintains an exclusive API arrangement with OpenAI through 2030 unless an earlier artificial general intelligence determination applies, the pricing shift carries significant implications. Cheaper inference can stimulate demand for Azure-hosted AI services, but it also compresses revenue per token. The tension between volume growth and margin preservation is central to Microsoft's cloud narrative.
A Three-Tier Pricing Structure
OpenAI's GPT-6 family now spans a wide cost spectrum. At the high end, Astra charges $10 per million input tokens and $50 per million output tokens, translating to a combined cost of $60 for one million tokens in each direction. Sol, by contrast, costs $12 combined, while Luna sits at just $0.60 combined — 99% below Astra. Sol's positioning is deliberate: it costs 20 times more than Luna but only one-fifth as much as Astra. OpenAI markets Sol for complex coding and agentic workloads where token consumption tends to be heavy.
All three models share a 1.05-million-token context window and a 128,000-token output limit. Sol supports reasoning settings ranging from none to maximum. OpenAI has not disclosed a launch date on its reviewed pages, making the current listing the only verified development.
Microsoft Shares Slip on Quiet Volume
Microsoft shares closed at $498.00 on September 22, down 0.72% from the prior close of $501.61. The stock opened higher at $507.32 and reached a session high of $508.50 before reversing to a low of $493.65. Volume totaled 14.99 million shares, roughly 32% below the three-month average of 22.18 million. The subdued trading activity weakens any argument that the GPT-6 Sol announcement directly triggered the decline. No public disclosure tied the move to the pricing update.
Microsoft shares remain 10.1% below their 52-week high and trade at 27.75 times reported earnings. The valuation reflects investor expectations for continued cloud growth, but also caution about the cost of delivering it.
The Cloud Margin Squeeze
Microsoft's fiscal fourth-quarter results, reported July 29, illustrate the challenge. Cloud revenue climbed 27% to $59.3 billion, with Azure growing 43%. However, cloud gross margin fell to 65% as AI infrastructure and usage costs rose. Capital spending reached $41 billion in the quarter, while free cash flow was $19.6 billion. CFO Amy Hood noted during the earnings call that "demand continues to exceed available supply."
Sol's lower unit prices could expand API traffic and Azure consumption, but they may also reduce revenue per token before utilization or efficiency improvements catch up. Neither OpenAI nor Microsoft disclosed Sol's inference cost or expected volume, leaving the margin impact uncertain.
Analyst Views Diverge
Wall Street remains split on Microsoft's prospects. Recent recommendations show a wide range of targets. Timothy Horan of Oppenheimer reiterated a Buy rating with a $570 target, implying 14.5% upside. Raimo Lenschow of Barclays maintained a Buy with a $512 target (2.8% upside). Thomas Blakey of Cantor Fitzgerald kept a Buy and a $608 target (22.1% upside). Brent Thill of Jefferies also maintained a Buy with a $575 target (15.5% upside). The lone Hold came from Alexander Haissl of Rothschild Redburn, who reiterated a Hold with a $440 target, suggesting 11.6% downside.
The $168 spread between the highest and lowest targets equals 33.7% of Microsoft's closing price. Bulls are betting on sustained cloud momentum, while the cautious view allows for weaker returns on heavy AI spending.
The Cannibalization Risk
The strongest counterargument to the bullish thesis is cannibalization. Developers may migrate from Astra to the cheaper Sol without increasing total token consumption enough to offset the price decline. Sol's workload quality also remains untested in disclosed customer data.
Risks include the possibility that Microsoft gains Azure volume while cloud margin continues to fall if Sol usage proves compute-heavy, heavily discounted, or displaces higher-priced models faster than infrastructure efficiency improves. Microsoft has not announced its next earnings date, but the next filing must answer a concrete question: did cheaper model access lift Azure consumption faster than it compressed cloud margin?
