Gemini 3.8 Flash is live: coding-tuned, in the API at $0.75/M today
Google shipped its workhorse to the API, Antigravity and Android Studio at once — it beats bigger models on DeepSWE, and the intro price doubles Jan 1.
Copy markdown
It's live now, everywhere you build
Gemini 3.8 Flash shipped today in the Gemini API and AI Studio, plus Vertex/Gemini Enterprise, Google Antigravity, Android Studio and Stitch — no waitlist. Google frames it as the workhorse for long-horizon coding and autonomous agents, not a chat toy.
It punches above its price on coding
On DeepSWE v1.1 (long-horizon software engineering) Google says 3.8 Flash beats most larger frontier models, and it scores 54.9% on HLE-Verified for multi-step reasoning. If you're running agent loops on a pricier Pro-tier model, this is the swap worth benchmarking this week.
$0.75/$3.75 per M — but the clock runs to Dec 31
Intro pricing is $0.75 per M input and $3.75 per M output through December 31; on January 1, 2027 it doubles to $1.50/$7.50. Size your agent budgets against the post-intro rate, because that's what you'll actually pay in a few months.
A gated 'Cyber' variant that finds real bugs
Gemini 3.8 Flash Cyber replaces 3.5 Cyber, reporting a 70%+ real-world vulnerability-detection rate across 20 languages and 2.6x more correct patches than leading models on Chrome Security. You can't just call it, though — it's limited to vetted researchers, governments and critical-infrastructure operators via the new Fairwind Program.
Check the cutoff before you trust it
Knowledge is uneven: Google cites a March 2026 cutoff for some domains but as far back as January 2025 for others, and it hasn't confirmed the context-window size. Lean on it for code and agent orchestration; verify anything time-sensitive it hands you.
Elsewhere: Google's TimesFM-3 tops forecasting leaderboards
Google also released TimesFM-3, a 330M-param zero-shot foundation model for multivariate time-series forecasting that leads three public leaderboards. The catch: the weights are non-commercial, so it's for prototypes and research — not something you can ship into a product yet.