Gemini 3.7 Flash Is Official: $0.75/1M Input, 1M Context, Confirmed Pricing

Update — August 14, 2026: Google officially launched Gemini 3.7 Flash via a blog.google post from Tulsee Doshi, Senior Director of Product Management. The August 13 leak this post originally covered turned out to be fully accurate on pricing: $0.75 per million input tokens and $3.75 per million output tokens, exactly as claimed. The Gemini 3.5 Pro "cancellation" and the Sergey Brin recursive-self-improvement claim were not part of Google's official announcement and remain unconfirmed. This post has been rewritten to lead with what's now confirmed, with the leak's timeline preserved below for reference.

Gemini 3.7 Flash is real. Google's official post — corroborated by X posts from Logan Kilpatrick, Google DeepMind, and Google AI Studio on August 13-14, 2026 — calls it Google's "most intelligent workhorse model yet for coding and agents." It ships at $0.75 per million input tokens and $3.75 per million output tokens, an introductory price through December 31, 2026, with a 1 million token context window, multimodal support, and availability across Antigravity, the Gemini API via AI Studio, Android Studio, the Gemini Enterprise Agent Platform, the Gemini Enterprise app, and Gemini Spark. CEO Sundar Pichai called it a "workhorse for performance at great value."

TL;DR — what's confirmed vs. what's still a rumor

What Google actually confirmed on launch day

From the official blog.google announcement, plus the DeepMind and AI Studio posts backing it up:

• Model: Gemini 3.7 Flash, described as the "most intelligent workhorse model yet for coding and agents"
• Pricing: $0.75 per million input tokens / $3.75 per million output tokens, introductory through December 31, 2026, then $1.50/$7.50 starting January 1, 2027 — Google also cut Gemini 3.6 Flash's price to match
• Context window: 1 million tokens, multimodal
• Availability: Antigravity, the Gemini API via AI Studio, Android Studio, the Gemini Enterprise Agent Platform, the Gemini Enterprise app, and Gemini Spark, live now
• Powers Gemini Spark: a 24/7 personal agent for Google AI Pro/Ultra subscribers, rolling out today in 160+ countries
• Safety: Google says it updated Frontier Safety Framework safeguards for CBRN and cyber-offense risk domains alongside this release
• Cadence: ~3 weeks after Gemini 3.6 Flash's July 21, 2026 launch — Logan Kilpatrick attributed the jump to "algorithmic improvements," not just more compute or data
• Quote: Sundar Pichai called it a "workhorse for performance at great value"

That three-week cadence matters as much as the price. explainx.ai has now tracked three Flash-tier releases from Google in under two months — 3.5 Flash-Lite and 3.6 Flash on July 21, and now 3.7 Flash on August 14 — while Gemini 3.5 Pro remains stuck at "testing with partners."

Benchmarks: where Gemini 3.7 Flash leads, and where it doesn't

Google DeepMind published these charts alongside the launch (methodology at deepmind.google/models/evals-methodology/gemini-3-7-flash). Four takeaways, kept honest about where it doesn't win:

• The jump over Gemini 3.6 Flash is the real story. Gemini 3.7 Flash gains 16.7 points on DeepSWE V1.1 and 13.4 points on AutomationBench over its own predecessor in three weeks — a bigger single-generation jump than 3.6 Flash's 17% output-token efficiency gain delivered against 3.5 Flash.
• It's not a universal win. GPT-5.6 Terra still leads DeepSWE V1.1 by more than 4 points and OSWorld-2.0 (computer use) by over 12 points — Gemini 3.7 Flash's edge shows up in enterprise automation and web dev, not long-horizon coding or agentic computer control.
• Muse Spark 1.2 leads on enterprise task quality. GDPVal-AA v2 Elo has Muse Spark 1.2 at 1628 and Claude Sonnet 5 at 1598, both ahead of Gemini 3.7 Flash's 1525 — a reminder this is a strong generalist release, not a frontier-beating one across the board.
• The margins against Claude Sonnet 5 on production code quality are narrow. FrontierCode 1.1 Main has Gemini 3.7 Flash at 43.6% vs. Claude Sonnet 5's 42.7% — under a point apart, well inside the range where a private eval harness matters more than the headline number.

What the leak got right — and what it didn't

This is worth being precise about, because a confirmed launch does not retroactively validate every claim in a leak that preceded it. Here's the August 13, 2026 leak chain — from accounts including @synthwavedd, @LuminaBench, @AiBattle_, and @cheatyyyy — scored against what Google actually announced:

So the pricing half of the leak was fully accurate — a genuinely rare outcome for an unsourced X thread. But the lesson from the original coverage still holds: a rumor can bundle a true claim with unrelated, unverified ones, and getting the numbers right doesn't automatically clear the rest. Treat the 3.5 Pro and Brin/RSI claims exactly as skeptically today as before this launch — nothing about Gemini 3.7 Flash's release confirms or denies either.

For context, this is the same pattern explainx.ai flagged in the July 13 Gemini 3.5 Pro benchmark leak, which called a July 17 launch that instead slipped into "testing with partners" status by the time Gemini 3.6 Flash actually shipped on July 21. Leak chains around Gemini have now gone 1-for-2 on hitting their claimed launch dates, but this time landed the pricing exactly.

How Hacker News reacted

The launch pulled a 662-point, 376-comment thread on Hacker News — mixed, not celebratory. A few threads worth knowing before you route production traffic:

• Speed over raw intelligence. Several commenters said Gemini 3.7 Flash's real edge is inference latency, not benchmark scores — one described it as the model "you reach for every time you do a Google search."
• Google's API friction is a recurring complaint, not a new one. Developers flagged that Google Cloud API key and billing setup remains higher-friction than OpenAI's or Anthropic's onboarding — a long-standing gripe resurfacing on this thread.
• Skepticism about "introductory pricing through Dec 31, 2026." Commenters questioned how meaningful that window is given how fast frontier models get superseded.
• One independent design-fidelity test cut against expectations. An image-to-HTML comparison (posted by HN user jjcm) found Claude Opus 5 still best-in-class for design fidelity, but Gemini 3.7 Flash notably ahead of Grok 4.6 on the same task.
• "Still no Gemini 3.5 Pro" fatigue. With three Flash-tier releases in under two months and no Pro-tier shipment, some commenters questioned whether DeepMind is still operating as a frontier lab on the Pro tier.

What this means for builders

• The $0.75/$3.75 pricing plus 1M context is worth testing now on high-volume workloads — document extraction, agentic search, and classification jobs are exactly where a Flash-tier price cut and larger context window compound. It's live today across Antigravity, AI Studio, Android Studio, Gemini Enterprise, and Spark, so there's no wait.
• Budget for the January 1, 2027 price jump. The $0.75/$3.75 rate is introductory only, through December 31, 2026 — it reverts to $1.50/$7.50 after that. Don't build a cost model that assumes today's price is permanent.
• Re-run your own harness before switching production traffic. Independent testing of Gemini 3.6 Flash found some workloads got less token-efficient despite lower sticker prices — the same caution applies here, and Hacker News commenters are already flagging that latency, not raw benchmark score, is where this model actually differentiates. explainx.ai's token-budget planning guide covers how to set that harness up.
• Route by task, not by headline benchmark. GPT-5.6 Terra leads long-horizon software engineering and computer use; Muse Spark 1.2 leads enterprise task quality; Gemini 3.7 Flash leads enterprise automation and web dev; Claude Sonnet 5 stays competitive on production code quality within a point. Pick per-task, not per-vendor.
• Watch for the 3.5 Pro and Brin/RSI threads separately. Neither was addressed by this launch — if you were waiting on either to be confirmed, keep watching blog.google rather than assuming this release settles them.

Related reading

• Gemini 3.7 Flash vs Grok 4.6 vs Sonnet 5 vs GPT-5.6: the real numbers — full benchmark comparison from Google's own launch charts
• Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber: what actually changed
• Gemini 3.5 Pro benchmark leak — July 13, 2026
• Google's Frozen v2 chip and the start of Gemini 4 pre-training
• Gemini hit 1 billion users — but it's not the same billion as ChatGPT
• Weco AIDE² — Level 1 recursive self-improvement, 8 days, 7 agent versions
• DeepMind's AGI-to-ASI paper: four pathways
• Google Gemini 3.5 complete guide
• Official: Google's Gemini 3.7 Flash launch post · Google Gemini API · 9to5Google's Gemini 3.6 Flash launch report

Sources: Google's official blog.google launch post by Tulsee Doshi (Senior Director, Product Management), August 13-14, 2026; corroborating X posts from Google, Logan Kilpatrick, Google DeepMind, and Google AI Studio; DeepMind evals methodology at deepmind.google/models/evals-methodology/gemini-3-7-flash; Simon Willison's pricing-page diff against an August 9, 2026 archive.org snapshot; Hacker News discussion (662 points, 376 comments). Original leak sourced from X posts by @synthwavedd, @LuminaBench, @AiBattle_, and @cheatyyyy, August 13, 2026.

Confirmed details (launch date, pricing, context window, availability, benchmark scores) reflect Google's official announcements as of August 14, 2026. The Gemini 3.5 Pro cancellation claim and the Sergey Brin recursive-self-improvement claim remain unverified by Google as of this update — this article does not treat them as confirmed just because the pricing leak they traveled alongside turned out to be accurate.