On August 13, a double blow was dealt to the artificial intelligence market: Google announced the Gemini 3.7 Flash model, and OpenAI introduced the Ultrafast mode for GPT-5.6 Sol. Both developments target one key market need — radically accelerating tasks related to programming and autonomous AI agents.
What's new in Gemini 3.7 Flash
Google positions Gemini 3.7 Flash as a workhorse for programming and business process automation. The company claims significant progress in multi-step planning, precise instruction following, and code generation quality. Notably, the cost of using the new model is halved compared to the starting price of its predecessor — an aggressive move in the price war for developers.

The Flash line is designed for scenarios where not only accuracy but also response speed is critical: coding, AI agent work, and tasks with many sequential calls to the model. This is a direct response to enterprise client demands, where delays in milliseconds translate into real money.
Let me remind you that in July, Google released Gemini 3.6 Flash with a price of $1.5 per 1 million input tokens and $7.5 per 1 million output tokens. At that time, the model showed 17% lower output token consumption compared to 3.5 Flash. However, the flagship Gemini 3.5 Pro has still not been released — developers are testing it with partners, and release timelines remain unclear. This creates intrigue: perhaps the company is revising its architecture in favor of speed at the expense of versatility.
How OpenAI accelerated GPT-5.6 Sol
OpenAI responded with the Ultrafast mode for GPT-5.6 Sol. The stated figures are impressive: up to 14 times faster than the standard mode and generation of up to 750 output tokens per second. The secret lies in the use of Cerebras infrastructure, which specializes in ultra-fast chips for AI.
At launch, Ultrafast is available via API to a limited circle of clients. OpenAI plans to gradually connect more users as computing capacity grows. I should note that the figure of 750 tokens per second was already mentioned during the GPT-5.6 announcement in June, but at that time the company honestly warned that the accelerated version would not be available to everyone at once.
This race for speed is not just marketing. In August, SpaceXAI also introduced Grok 4.6 with an emphasis on long-term agent tasks and programming. The market is clearly moving toward output token speed becoming a new battlefield, comparable in significance to the quality of the models themselves.
My analysis: Google's halving of prices and OpenAI's 14-fold acceleration are not just competition, but a fundamental shift in the AI economy. Fast and cheap models make mass agent scenarios profitable, which were previously unprofitable. Investors and developers should closely watch how these changes will affect the cost of computing resources and cloud providers' margins in the coming quarters.