Google halves the price of its newest coding model — AI development costs keep falling, fast
Google released Gemini 3.7 Flash on 13 August 2026 at a 50% introductory price cut with meaningfully better coding benchmarks than its predecessor from just three weeks earlier, the latest sign that the cost of AI-assisted software development is dropping faster than most buyers realise.
17 August 2026
Google DeepMind released Gemini 3.7 Flash on 13 August 2026, pitched explicitly as its “most intelligent workhorse model yet for coding and agents.” It’s priced at an introductory $0.75 per million input tokens and $3.75 per million output tokens through the end of 2026 — roughly half the launch rate of Gemini 3.6 Flash, which shipped only three weeks earlier. The coding numbers moved a similar amount in that short window: 43.6% on the FrontierCode 1.1 benchmark versus 34.4% for its predecessor, and 65.3% on DeepSWE versus 49.0%. The model is available immediately through the Gemini API, Google AI Studio, Android Studio and Google’s enterprise products.
Three-week release cycles with double-digit benchmark jumps and 50% price cuts are now the normal cadence among the major model providers, not an exception. Anthropic, OpenAI and Google are all competing on the same two axes at once — coding capability and per-token cost — which matters directly to anyone commissioning software: the compute cost of AI-assisted development keeps falling even as the tools get more capable, faster than most non-technical buyers are tracking. A quote based on last quarter’s model economics is already out of date.
So what
If you were quoted a build cost some months ago and shelved the project as too expensive, it’s worth asking again — the underlying AI-assisted delivery cost has likely moved since. This doesn’t mean AI makes software free or that quality corners can be cut; it means the economics of building properly, with AI assistance used well, keep improving. See our custom software development approach or get in touch for a current-day estimate rather than one based on stale assumptions.