# Google Halves Gemini Flash Pricing Three Weeks After Its Last Release

Gemini 3.7 Flash lists at $0.75 per million input tokens through the end of 2026 and reports a jump from 49.0% to 65.3% on the DeepSWE coding benchmark.

- Published: 2026-08-14T06:27:18.545Z
- Canonical: https://polylog.news/ai/2026-08-14/google-halves-gemini-flash-pricing-three-weeks-after-its-las
- Publisher: Polylog (AI desk)
- Section: tech
- Sources: [Google](https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-3-7-flash/), [Hacker News](https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-3-7-flash/), [Polylog editors](https://polylog.news)

Google released [Gemini 3.7 Flash](https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-3-7-flash/) on Thursday, updating the tier that carries most production traffic for developers on its platform. The introductory price runs at $0.75 per million input tokens and $3.75 per million output tokens through December 31, 2026, which is half the launch price of Gemini 3.6 Flash. From January 1, 2027, the rate rises to $1.50 and $7.50, [according to the pricing detail reported by VentureBeat](https://venturebeat.com/technology/googles-gemini-3-7-flash-targets-coding-and-agents-with-a-50-introductory-price-cut). The predecessor is three weeks old.

The main capability claim is about coding performance. Google reports its DeepSWE benchmark score rising from 49.0% to 65.3%, with Terminal-bench 2.1 at 85.8%. Those are launch figures from Google itself, not independently reproduced. Independent measurement so far is thinner. Artificial Analysis places [Gemini 3.7 Flash at 56 on its composite intelligence index](https://artificialanalysis.ai/models/gemini-3-7-flash). On Agent's Last Exam, a multimodal desktop and operating-system task set, Claude Sonnet 5 passes 33.3% of tasks against 26.3% for the new Flash model. OpenAI's GPT-5.6 Terra still leads on Terminal-bench 3.0 and OSWorld-2.0, [as the-decoder noted in its comparison](https://the-decoder.com/gemini-3-7-flash-lands-with-coding-gains-and-undercuts-its-three-week-old-predecessors-price-by-50/).

The Russian-language technical channel AI ML Big Data [described the release primarily as a price event](https://t.me/ai_machinelearning_big_data/10703), noting that Google updated its high-volume production model, not its top tier, and cut the rate in half on the tier where token volume is largest. That is the right way to read it for anyone running an agent loop, where output tokens make up most of the bill.

Gemini 3.7 Flash also becomes the model that runs Gemini Spark, Google's background personal agent, which [runs on dedicated virtual machines and connects to Gmail, Docs and Sheets](https://techcrunch.com/2026/05/19/google-introduces-gemini-spark-a-24-7-agentic-assistant-with-gmail-integration/). A cheaper, more reliable tool-calling model is exactly what Spark needed, because an agent that retries less often costs less to run and fails less often in front of a user.

## What this means

Google is competing on the price of the token, not the maximum capability of the model. Cutting the price of its highest-volume tier in half three weeks after the predecessor's launch increases competitive pressure on every vendor selling mid-tier inference at a margin, including Anthropic's Sonnet line, OpenAI's Terra and Luna variants, and the Chinese open-weight providers whose pitch is price. Inference resellers and agent startups whose unit economics assume a stable price per token are exposed to this pressure, while application builders benefit because their gross margin improves without any engineering work. The introductory rate expires December 31, so the real test is whether Google keeps the permanent price at this level or extends the discount to hold market share.

## What to watch

- Whether independent evaluators reproduce the DeepSWE jump from 49.0% to 65.3%, since a vendor-reported coding gain of that size usually shrinks under third-party harnesses.
- Whether OpenAI or Anthropic responds with a price cut on their own mid-tier models, which would confirm that repricing, not capability, is now the main competitive move.
- What happens to the price on January 1, 2027, because extending the discount would signal Google values share over margin on inference.