Gemini 3.5 Flash: Faster AND Actually Useful
At Google I/O on May 19, Google DeepMind officially released Gemini 3.5 Flash — the first model in the Gemini 3.5 family. Its tagline is "frontier intelligence with action," meaning it doesn't just answer questions; it executes real tasks.
The announcement was co-signed by Koray Kavukcuoglu (CTO of Google DeepMind), Jeff Dean (Chief Scientist), Oriol Vinyals, and Noam Shazeer — a lineup that signals how important this release is.
Speed meets intelligence
3.5 Flash runs at Flash-series speeds while rivaling flagship models on intelligence.
Key benchmarks released by Google:
- Terminal-Bench 2.1 (terminal task automation): 76.2%
- GDPval-AA (code generation): 1656 Elo
- MCP Atlas (tool calling): 83.6%
- CharXiv Reasoning (multimodal reasoning): 84.2%
All metrics surpass the previous-gen Gemini 3.1 Pro. Google also claims output is 4x faster than other frontier models.
An AI that actually does things
The biggest change is in "action." 3.5 Flash natively supports tool calling and agentic workflows — writing code, running terminals, calling APIs, processing data — all without step-by-step human guidance.
Google launched Antigravity, a new agent development platform, alongside 3.5 Flash. Developers can access the model through Google AI Studio and Android Studio.
The context window is 1M tokens, handling long documents and large codebases without issues.
Pricing
Standard pricing: $0.75/1M input tokens, $4.50/1M output tokens. Thinking mode doubles that to $1.50/$9.00.
This sparked discussion on HN — "Flash priced like Pro." But $4.50 for output includes thinking tokens, which is competitive against GPT-4o at $10/1M output.
Available now
3.5 Flash is live in the Gemini app, Google Search AI Mode, Google AI Studio, and Android Studio. Enterprise users can access it through Gemini Enterprise and Enterprise Agent Platform.
Google also revealed that 3.5 Pro is already in internal use and will launch next month.
HN reaction
The HN post hit 589 points with 68 comments. Reception was mixed:
- Supporters praised benchmark scores, speed, and value
- Skeptics noted benchmarks might be cherry-picked, avoiding areas where Claude excels
- Others acknowledged the price hike but admitted the model delivers
Bottom line: 3.5 Flash is one of the best value agentic models available today. It far exceeds expectations for a Flash-series upgrade, especially for workflow automation and tool calling. Now we wait to see if 3.5 Pro can push reasoning depth even further.
