Photo by Lukenn Sabellano on Unsplash
- As of May 19, 2026, Gemini 3.5 Flash processes 289 tokens per second — 3.6× faster than Gemini 3.1 Pro — and now outperforms that older flagship on all three major agentic benchmarks.
- API pricing stands at $1.50 per 1M input tokens and $9.00 per 1M output tokens (as of July 2, 2026, per Google's AI Blog): 25% cheaper than Gemini 3.1 Pro, but roughly 3× the Gemini 3 Flash Preview it replaced.
- Android 17 shipped June 16, 2026, first to Pixel 6 series and newer; Google's Android chief formally reframed the OS as an "intelligence system" with Gemini Intelligence at its core.
- Gemini 3.5 Pro remains in limited enterprise preview as of early July 2026, per TechCrunch — its public release has already slipped once from June to July.
What Happened
What if the budget-tier model in an AI family stopped being the compromise option and quietly became the best performer across the board? On May 19, 2026, that is exactly what Google announced at I/O 2026 — held at Shoreline Amphitheatre in Mountain View, California.
According to blockchain.news, which aggregated Google's June 2026 product releases with particular focus on the multimodal and translation capabilities, the company unveiled Gemini 3.5 Flash at the conference alongside a sweeping reframe of its entire AI strategy. CEO Sundar Pichai declared the start of what he called the "agentic Gemini era," describing it as "a moment where AI systems move decisively from answering questions to independently planning, reasoning, and completing complex multi-step tasks." Google's official AI Blog added the more pointed observation: Gemini 3.5 Flash "broke one of the unwritten rules of AI releases: the cheap, fast Flash tier now outperforms the previous flagship Pro model on coding and agentic benchmarks."
The technical baseline matters here. Gemini 3.5 Flash launched May 19, 2026 with a 1,048,576-token input context window and 65,536 maximum output tokens, supporting text, image, audio, and video inputs natively. Three weeks later, on June 16, 2026, Android 17 began rolling out — first to Pixel 6 series devices and newer — with Gemini Intelligence embedded directly at the operating system layer. Gemma 4 12B also arrived, a local model running on just 16GB of memory with vision and native voice processing, offering teams an on-device path for workflows where cloud routing is a liability.
Why the Benchmark Flip Matters for Agentic Workflows
The throughput number is where to start. As of July 2, 2026, according to Google's official AI Blog, Gemini 3.5 Flash runs at 289 tokens per second compared to approximately 80 tok/s for Gemini 3.1 Pro — a 3.6× speed improvement. In production agentic systems where a single user request can trigger a cascade of sequential model calls, that gap compounds into dramatically lower end-to-end latency. For teams building AI investing tools, financial planning automation, or multi-step document pipelines, faster inference is not a benchmark footnote — it is the difference between a viable product and a slow one.
The benchmark results reinforce the throughput story. As of July 2, 2026, Gemini 3.5 Flash scores 76.2% on Terminal-Bench 2.1 against Gemini 3.1 Pro's 70.3%, 83.6% on MCP Atlas versus 78.2%, and 57.9% on Finance Agent v2 compared to 43.0% — a 14.9-point gap on the benchmark most directly relevant to investment portfolio automation and enterprise financial workflows. This broader shift toward autonomous AI — which the team at AI Agents for Business examined in depth — is now the explicit design goal, not a secondary capability layer.
Chart: Gemini 3.5 Flash vs. Gemini 3.1 Pro on three agentic benchmarks, per Google's AI Blog (May 2026). Higher is better. Finance Agent v2 shows the largest gap at 14.9 percentage points.
Two supporting features round out the release. Gemini 3.5 Live Translate, as of July 2, 2026, handles real-time speech-to-speech translation across 70 languages while preserving the speaker's natural intonation and eliminating awkward pauses — a concrete unlock for multinational financial services teams that have historically patched together third-party translation layers. Antigravity 2.0, Google's upgraded agent-first development platform, now runs on Gemini 3.5 Flash and supports orchestration of parallel multi-agent workflows for long-horizon enterprise automation. Google also announced the $2 million Build with Gemini XPRIZE Hackathon — the largest prize pool ever for a hackathon — to push developer adoption of agentic workflows into production faster.
The Price Tag Hidden in the Announcement
Here is where the divergence between outlets becomes operationally meaningful. Blockchain.news and Google's own communications led with the performance story. XDA Developers ran a distinctly different headline: "Gemini 3.5 Flash costs 3x the model it replaced, and the era of cheap AI is ending." Both framings are accurate. Which one matters more depends entirely on your current stack's baseline.
As of July 2, 2026, Gemini 3.5 Flash API pricing sits at $1.50 per 1 million input tokens and $9.00 per 1 million output tokens. Compared to Gemini 3.1 Pro at $2.00 per million input and $12.00 per million output, that is a clean 25% reduction — a real benefit for any team currently paying Pro-tier rates. But for integrations running on Gemini 3 Flash Preview at roughly $0.50 per million tokens, the new Flash model represents approximately a 3× cost increase. The API limit math looks very different depending on which baseline you are measuring from, and high-volume workloads — stock market today monitoring, real-time compliance scanning, automated research pipelines — will feel that delta immediately in production budgets.
TechCrunch reported a detail other outlets largely skipped: Gemini 3.5 Pro "still wasn't out as of mid-June 2026, with the June target slipping to July, remaining in limited preview for enterprise customers." If your roadmap depends on Pro-tier capabilities, that timeline has already moved once. Plan accordingly.
My read on the pricing situation: for teams that built high-volume integrations at the old Flash Preview price point, the cost recalculation is the more urgent operational decision — more so than any benchmark delta. Benchmark improvements are durable; quarterly budgets are not.
Android 17, Three Steps, and Who Should Wait
Android 17's June 16, 2026 launch changed the framing of the OS category itself. Google's Android chief described the platform at I/O 2026 as an "intelligence system" rather than an operating system, with Gemini Intelligence as its first concrete expression of that shift. For Pixel 6 and newer users, those capabilities are already live. For everyone else on Android, broader device support is rolling out throughout 2026 — specific timelines depend on each OEM's update cadence, which varies considerably across manufacturers.
Three concrete actions worth taking before the end of the week:
If your stack calls the Flash tier at volume for use cases like real-time financial planning alerts or high-throughput document classification, calculate the full cost delta first. A roughly 3× input token price increase at production scale is a budget line item that needs a sign-off, not a rounding error absorbed quietly.
The 57.9% versus 43.0% Finance Agent v2 gap is the most compelling data point for teams building investment portfolio automation or AI investing tools on Gemini. But benchmark averages mask significant task-specific variance. Run your actual production workflow against both tiers on a representative test batch before committing your architecture to Gemini 3.5 Flash.
TechCrunch flagged the Pro tier delay — the June target slipped to July in limited enterprise preview. Do not architect production systems around a preview API that has already missed one target date. Gemma 4 12B, running locally on 16GB of memory with on-device voice and vision, may be the smarter interim option for latency-sensitive or data-sensitive workflows that cannot afford to wait on a moving GA window.
Frequently Asked Questions
What is Gemini 3.5 Flash and how does it differ from previous Flash models?
Gemini 3.5 Flash is Google's latest efficient AI model, launched May 19, 2026. Unlike prior Flash tiers that traded capability for cost, Gemini 3.5 Flash now outperforms the older Gemini 3.1 Pro flagship on three agentic benchmarks: Terminal-Bench 2.1 (76.2% vs. 70.3%), MCP Atlas (83.6% vs. 78.2%), and Finance Agent v2 (57.9% vs. 43.0%). It processes 289 tokens per second, supports a 1,048,576-token context window, and handles text, image, audio, and video inputs natively. The key tradeoff versus older Flash models: at $1.50 per 1M input tokens as of July 2, 2026, it costs roughly 3× the Gemini 3 Flash Preview tier it replaced.
How much does Gemini 3.5 Flash cost per 1 million tokens versus Gemini 3.1 Pro?
As of July 2, 2026, per Google's official AI Blog, Gemini 3.5 Flash is priced at $1.50 per 1 million input tokens and $9.00 per 1 million output tokens. Gemini 3.1 Pro costs $2.00 per 1 million input tokens and $12.00 per 1 million output tokens — making Flash 25% cheaper on both dimensions for teams migrating down from Pro. However, teams currently on Gemini 3 Flash Preview at roughly $0.50 per million tokens face approximately a 3× price increase. Run your specific API limit math before assuming the new model is cost-neutral for your workflow.
When will Android 17 AI features reach non-Pixel Android phones?
Android 17 began rolling out June 16, 2026, starting with Pixel 6 series and newer devices. Broader support across other Android manufacturers is expected throughout the remainder of 2026, but specific dates depend on each OEM's own update cycle. As of July 2, 2026, no manufacturer-specific rollout timelines have been publicly confirmed. Samsung, OnePlus, and similar makers typically follow Pixel launches by several months under their own software schedules.
Disclaimer: This article is editorial commentary based on publicly available reporting and does not constitute financial or investment advice. No independent product testing was conducted by this publication. Research based on publicly available sources current as of July 2, 2026.