OpenAI Releases GPT-6.2 Astra and Sol: Flagship Model Prices Cut by 20%, Further Lowering Usage Costs
On November 17, OpenAI released GPT-6.2 Astra and GPT-6.2 Sol. Astra features a 20% unit price reduction alongside improved professional task performance and token efficiency, while Sol offers near-Astra coding capabilities at approximately one-fifth of the cost. For OpenAI, revenue growth relies on whether incremental paid usage driven by these price cuts can offset lower unit prices. Investors should monitor whether volume expansion offsets unit price declines and track OpenAI's compute costs for task completion.

TradingKey - On November 17, OpenAI released GPT-6.2 Astra and GPT-6.2 Sol. The flagship model Astra achieved higher scores in professional task benchmarks, with standard API input and output unit prices both cut by 20% compared to the previous generation; meanwhile, Sol, whose pricing remained unchanged, delivered performance close to Astra in a coding benchmark at approximately one-fifth of the cost. Both models reduce the cost for users to complete certain AI tasks. For OpenAI, revenue growth will depend more on how much new usage the price cuts can generate.
Astra Boosts Professional Task Scores as Sol Achieves Close Coding Results at 20% Cost
Astra is primarily geared toward financial analysis, legal drafting, and complex programming. According to an announcement by OpenAI, in the GDPval v2 benchmark based on real-world professional tasks, the new Astra scored 6.2 percentage points higher than the prior generation, while using 22% fewer output tokens to generate answers. A token is the unit of measurement for API billing usage. In other words, on this set of tasks, users not only obtain output at a lower unit price, but also require fewer billable units to achieve results.
Sol's progress is reflected in code modifications. FrontierCode 2.0 evaluates whether modifications made by a model to real-world projects meet maintainer standards for accepting a merge. When both models operated at the highest reasoning effort, the score difference between Sol and Astra was no more than 0.5 points, with testing costs running at about one-fifth of Astra's. While this result is limited to specific programming tasks, it demonstrates that lower-cost models can now undertake a portion of the work that might previously have been assigned to flagship models.
Astra Cuts Unit Price by 20% as New Paid Volume Determines Post-Cut Revenue Performance
The standard price for Astra has dropped from $10 to $8 per 1 million regular input tokens, and from $50 to $40 per 1 million output tokens. If the ratio of inputs to outputs remains unchanged, usage that previously generated $100 in revenue now generates only $80. To bring this revenue back to $100, the total volume of billable tokens would need to grow by 25%. This calculation is specific to Astra's standard API business and is not a forecast for OpenAI's overall revenue.
For investors, the distinction between call volume and paid usage is particularly critical. If developers previously checked code only once prior to a software release, but switch to checking it after every modification following the price cut, OpenAI gains new paid demand. Conversely, if users continue to complete the same number of tasks but consume fewer tokens each time, or shift part of the workload to the cheaper Sol, actual spending may decline even if usage frequency does not drop.
Therefore, in evaluating the commercial impact of this release, the revenue side should focus on whether incremental paid usage can offset the decline in unit prices, while the profit side should focus on OpenAI's actual compute costs for completing similar tasks.
This content was translated using AI and reviewed for clarity. It is for informational purposes only.
Recommended Articles












Comments (0)
Click the $ button, enter the symbol, and select to link a stock, ETF, or other ticker.