Three weeks is not a product cycle. It is barely a sprint. But three weeks is exactly how long Gemini 3.6 Flash held the top of Google's workhorse lineup before the company replaced it — and then cut the price of the replacement in half.
Google DeepMind released Gemini 3.7 Flash on August 13, calling it "our most intelligent workhorse model yet for coding and agents." The model ships with an introductory price of $0.75 per million input tokens and $3.75 per million output tokens, available through December 31, 2026. It retains the 1-million-token context window of its predecessor, accepts text, image, video and speech input, and is live in the Gemini API, Google AI Studio, Android Studio, the Antigravity agent IDE, and the Gemini Enterprise Agent Platform. Google AI Pro and Ultra subscribers get it through Spark, the 24/7 personal agent Google launched at I/O.
The headline numbers are about code. On FrontierCode 1.1 Main, Cognition's benchmark for production-grade code quality, Gemini 3.7 Flash scores 43.6%, up from 34.4% for 3.6 Flash — enough to edge past Claude Sonnet 5 at 42.7% and GPT-5.6 Terra at 41.3%. On DeepSWE v1.1, a long-horizon software engineering eval, it jumps to 65.3% from 49.0%, though GPT-5.6 Terra still leads that board at 69.6%. On Arena.ai's WebDev Arena, 3.7 Flash posts an Elo of 1588 against 1538 for its predecessor, the top score in its class.
The enterprise pitch
The more striking gain is in workflow automation. On AutomationBench, Zapier's evaluation of real-world business process completion, Gemini 3.7 Flash scores 30.4% — nearly double 3.6 Flash's 17.0%, and well clear of GPT-5.6 Terra at 23.6% and Claude Sonnet 5 at 10.7%. Document comprehension follows the same curve: 34.0% on the GDP.pdf benchmark, up from 22.0%.
Writing on behalf of the Gemini team, Tulsee Doshi, senior director of product management, framed the fast turnaround as intentional. The release "comes just three weeks after Gemini 3.6 Flash, and is a direct result of developer feedback and algorithmic innovations that we look forward to bringing to future models," she wrote. Logan Kilpatrick, who leads product for AI Studio, credited "awesome algorithmic improvements" from teams across Google DeepMind for the compressed cadence.
Google's qualitative claims are more interesting than its benchmark table. The company says 3.7 Flash "better adapts to roadblocks, clarifies intent when needed, and follows instructions with greater fidelity," and that it "thinks more diligently." That is a description of an agent, not a chatbot.
It is not a clean sweep. Gemini 3.7 Flash trails badly on agentic computer use, scoring 38.1% on OSWorld-2.0 against GPT-5.6 Terra's 50.2%. Claude Sonnet 5 tops Agent's Last Exam at 33.3% to Gemini's 26.3%. And on GDPval-AA v2, the knowledge-work eval, Muse Spark 1.2 leads at 1628 Elo with Gemini 3.7 Flash trailing the field at 1525 — despite a 103-point improvement over 3.6 Flash.
Independent benchmarking from Artificial Analysis broadly corroborates Google's story while trimming the framing. The firm scored Gemini 3.7 Flash (high reasoning) at 56 on its Intelligence Index, a four-point gain over 3.6 Flash but still a point behind GPT-5.6 Terra and Muse Spark 1.2 at 57. What sets the model apart in that analysis is throughput: roughly 340 output tokens per second, nearly three times GPT-5.6 Terra's speed, translating to an average time-per-task of 1.7 minutes and a spot on the Intelligence vs. Time per Task Pareto frontier.
The math behind the price cut
One correction to the popular framing: the 50% cut is measured against list price, not against what anyone was actually paying. Artificial Analysis notes that Gemini 3.7 Flash retains 3.6 Flash's standard pricing of $1.50/$7.50 per million tokens — the discount is a promotion that expires with the calendar year, after which the sticker doubles. Measured in cost per Intelligence Index task, the real-world improvement is $0.40 versus 3.6 Flash, a 30% reduction rather than 50%. Cached input tokens keep their 90% discount.
That still leaves Google undercutting the field dramatically. GPT-5.6 Terra lists at $2.00 input and $12.00 output; Claude Sonnet 5 at $2.00 and $10.00. Even Muse Spark 1.2, at $1.25 and $4.25, is more expensive.
Analysis: the mid-tier is becoming a commodity
Strip away the model names and what remains is a price war in the tier where most production tokens get spent. Frontier models win demos. Mid-tier models win invoices — and the mid-tier is converging on capability while diverging sharply on cost.
"The base model layer is commoditizing," said Amit Chandak, chief analytics officer at Kanerika, arguing that differentiation is migrating to data readiness, governance and orchestration. "Token cost has been the practical ceiling on scaling AI beyond isolated pilots."
Sanchit Gogia, chief analyst at Greyhound Research, put the structural point more bluntly. "The more important development is the continued compression of the price of useful machine intelligence," he said. "Capability, latency and cost are becoming inseparable buying criteria." His verdict on the launch data was appropriately cautious: "These remain vendor benchmark claims until the new model accumulates sufficient independent production evidence."
The divergence is real: the same week Google halved Flash pricing, DeepSeek raised some V4 prices more than tenfold as demand strained capacity. Two vendors, opposite directions, same market.
What to watch
The obvious tripwire is January 1, 2027, when Gemini 3.7 Flash pricing reverts to $1.50/$7.50. Teams that build agent pipelines against $0.75 economics will face a doubling with no model change — a classic introductory-pricing trap, and a live question about whether Google intends to hold the discount or let it lapse.
Watch, too, for the Pro tier. Google has shipped three Flash models in three months while offering no timeline for its next Pro release; Sundar Pichai sidestepped the question on the last earnings call. A cadence that fast at the bottom and that slow at the top says something about where the returns currently are. Finally, watch OSWorld-2.0. Computer use is the weakest link in Google's agent story, and no amount of price cutting closes a 12-point gap.
“The more important development is the continued compression of the price of useful machine intelligence. Capability, latency and cost are becoming inseparable buying criteria.”— Sanchit Gogia, Chief Analyst, Greyhound Research