Google launches Gemini 3.7 Flash with higher coding scores and half-price introductory rates
- Gemini 3.7 Flash arrives three weeks after 3.6 Flash, with Google setting an introductory price at half the original 3.6 Flash cost per million tokens.
- On software engineering tests, FrontierCode 1.1 Main rises from 34.4% to 43.6% and DeepSWE v1.1 rises from 49.0% to 65.3% versus Gemini 3.6 Flash.
- For web development, WebDev Arena scores Gemini 3.7 Flash at 1588 Elo, compared with 1538 for 3.6 Flash; Google says it produces functional layouts and complete apps in fewer prompts.
- On document and workflow evaluations, GDP.pdf improves from 22.0% to 34.0%, while AutomationBench improves from 17.0% to 30.4% against 3.6 Flash.
- Google demonstrated Gemini 3.7 Flash with Nano Banana generating characters, items, and textures for a playable 3D game in real time from a text prompt.
Hacker News opinions
Pricing through December 2026 does not mean Flash development stops until next year. Google shipped 3.7 only three weeks after 3.6, and Gemini 4 is reportedly near.
I do not see why I would pick 3.7 Flash when Terra is about half the price on FrontierCode and Grok 4.6 appears both better and cheaper. Is DeepMind still a frontier lab?
I still use 3.6 Flash for token speed, latency, uptime, and large-context work. Grok 4.6 scored lower on some of my internal tests or ran slower.
Per-token pricing misses how many tokens a model consumes. Artificial Analysis puts Grok 4.6 at $1,068 for its suite and Gemini 3.7 Flash at $485, so Gemini costs less than half as much in that test.
Flash 3.6 High ran about 10 times faster than Luna xhigh for my workload and got similar results.
I want another Gemini Pro model. Google seems to be concentrating on fast, moderately capable models.
The model card puts it around GPT 5.6 Terra and Sonnet 5 depending on the benchmark. People cannot dismiss Google for releasing a model in the same segment that OpenAI and Anthropic also sell.
I think this is competitive: it beats Claude Sonnet 5 on most benchmarks and costs less than half as much. Google is not leading maximum capability, but it remains competitive on capability, price, and speed together.
For text-only work, DS V4 Flash and Pro look 13 to 26 times cheaper at similar intelligence, while Luna is about eight times cheaper. Gemini's clear advantage looks like speed, likely from TPUs.
Speed matters for my non-coding app. Users do not expect an LLM in this flow, so waiting seconds or minutes is unacceptable.
At high volume, Gemini Flash is the only provider I have found reliable enough to ingest 1 million documents per hour. DeepSeek, Luna, and Mistral failed roughly one in three requests for me.
Artificial Analysis has 3.7 Flash at 56, up from 52 for 3.6, but 3.7 uses 37,000 output tokens per task versus 26,000. Its halved token price still makes it cheaper per task.