Gemini 3.7 Flash — 50% cheaper than 3.6, ~3 weeks between releases

TL;DR

Logan Kilpatrick (Google AI Studio) announces Gemini 3.7 Flash, released roughly three weeks after 3.6 Flash. The pitch is speed plus a 50% price cut versus 3.6 Flash (promo through end of year) with an intelligence bump attributed to “algorithmic improvements.” Available in the Gemini API, AI Studio, Antigravity, and other Google surfaces. This is a closed-weights, API-only product update — no technical report, weights, or benchmark artifact accompanies the announcement.

Key claims

Gemini 3.7 Flash ships in the Gemini API, AI Studio, and Antigravity, positioned as faster than 3.6 Flash [tweet body].
Pricing is 50% lower than 3.6 Flash through end of year [tweet body].
Google attributes the intelligence gain over 3.6 Flash — landing in ~3 weeks — to algorithmic improvements rather than a scale-up [tweet body].
Four attached benchmark images are referenced but not transcribed here; the community reply flags them as “carefully chosen benchmarks and comparisons” [reply from @jkelleher].

Method

Product announcement. No architecture, training recipe, or eval methodology disclosed in the tweet or thread. The follow-up self-quote from Kilpatrick frames the 3.5 → 3.6 → 3.7 progression as a rapid iteration cadence across Google DeepMind, with an emphasis on “usability for real work.”

Results

Four benchmark charts are embedded (not machine-readable from the tweet text). No numeric claims beyond “50% cheaper” and “strong intelligence increase.” The 148K-view engagement and skeptical reply from a domain reader suggest the benchmark selection is not neutral; treat headline gains as vendor-reported until third-party evals appear.

Why it’s interesting

Sits inside a very tight release cadence for Google’s Flash tier: Build with Gemini 3 Flash, frontier intelligence that scales with you introduced Gemini 3 Flash in December 2025 at roughly 1/4 the cost of 3 Pro, and this drop is the third minor bump (3.5 → 3.6 → 3.7) since. The three-week gap between 3.6 and 3.7 is unusually short for a frontier-tier Flash model and matches the “algorithmic improvements over scale-up” framing also seen in GPT-5.3 Instant: Smoother, more useful everyday conversations and Introducing GPT-5.2 — frontier labs increasingly ship incremental post-training / distillation refreshes to closed API tiers between architecture-changing releases. Contrasts with the tech-report cadence of open releases (e.g. Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilities) where a Flash bump would usually come with a report card; here the artifact is the tweet.