Gemini 3.5 Flash Lands: What Builders Need to Know

A Flash That Beats Last Year's Pro

On 19 May 2026, Google unveiled Gemini 3.5 Flash at Google I/O and immediately made it the new default model across the consumer Gemini app, AI Mode in Google Search, and the Gemini API. The headline number is 76.2% on Terminal-Bench 2.1 — a benchmark built specifically to evaluate autonomous coding agents — against 70.3% for Gemini 3.1 Pro, the previous generation flagship. Google achieved that improvement while cutting the API price to $1.50 per million input tokens and $9.00 per million output tokens, roughly half the cost of Gemini 3.1 Pro at an equivalent performance tier.

What Gemini 3.5 Flash Actually Is

Flash is engineered for two primary use cases: high-volume AI agents and production coding tasks. It processes text, images, audio, and video as inputs and returns text output. Dynamic thinking is enabled by default, meaning the model allocates reasoning compute on a per-request basis — lighter for simple queries, deeper for complex ones — rather than requiring developers to choose a fixed mode at configuration time.

The context window is 1,048,576 input tokens with 65,536 output tokens. That input capacity is large enough to hold an entire medium-sized codebase in a single call, which is the architecture that agentic coding pipelines require. Cached input pricing drops to $0.15 per million tokens, rewarding architectures that aggressively reuse shared context such as system prompts and base documentation.

Benchmarks: The Numbers That Matter

On the agentic and coding suite, Flash beats Gemini 3.1 Pro across the board. Terminal-Bench 2.1 scores 76.2% versus 70.3% for Gemini 3.1 Pro. MCP Atlas, which measures tool-calling accuracy in multi-step agentic workflows, scores 83.6% versus 78.2%. CharXiv Reasoning, a scientific reasoning and chart analysis benchmark, scores 84.2% for Flash.

Gemini 3.1 Pro retains one meaningful edge: long-context retrieval above 128,000 tokens, where it holds a 7.6-point lead. For workloads that are pure document-retrieval tasks at extreme context lengths, Pro remains the better choice. For the majority of production workloads — coding pipelines, tool-calling agents, document summarisation, and conversational AI — Flash is now the stronger option.

Speed is the other critical dimension. Flash generates approximately 289 tokens per second, roughly four times faster than comparable frontier models from Anthropic and OpenAI in the same performance band. For real-time applications — streaming code completions, interactive agent sessions, live customer-facing chat — that throughput gap translates directly into user experience.

Pricing: The Economics of Switching

At $1.50 input and $9.00 output per million tokens, Gemini 3.5 Flash makes a clear economic case for high-volume workloads. Gemini 3.0 Ultra, Google's previous performance tier, cost approximately $3.50 per million input tokens. Teams that provisioned Gemini 3.1 Pro for coding or agent tasks now have the option to switch to Flash, receive better performance on agentic benchmarks, and pay less per million tokens simultaneously. That combination is unusual in the AI API market, where capability improvements typically arrive alongside price increases.

Gemini 3.5 Pro Is Coming in June

Sundar Pichai confirmed at Google I/O that Gemini 3.5 Pro is scheduled for general availability in June 2026. He told the audience directly: "Give us until next month to get it to you." Pro targets a 2-million-token context window — double Flash's capacity — along with a Deep Think reasoning mode for the hardest tasks: long-document synthesis, advanced scientific reasoning, and complex multi-file code refactoring. Pricing is expected at approximately $15 per million input tokens and $60 per million output tokens, consistent with Google's historical ten-to-one pricing ratio between Pro and Flash tiers.

Flash Now or Wait for Pro?

For most production workloads — coding agents, customer-facing chat, data extraction pipelines, and document automation — Flash is ready today. For workloads that genuinely require context beyond one million tokens, or that depend on maximum reasoning depth on every request, waiting two to four weeks for Pro is worth the delay. The decision is straightforward once you have defined your workload requirements.

What This Means for Indian Product Teams

Indian engineering teams evaluating AI infrastructure in June 2026 should treat Flash as the new baseline for Gemini integration. It is available today through Google AI Studio and the Gemini API, with Vertex AI integration for teams that require the enterprise contract, data-residency controls, and audit trails that cloud enterprise agreements provide.

The 289-token-per-second throughput matters especially for latency-sensitive consumer applications, where users on variable mobile connections in Tier 2 and Tier 3 cities experience the difference between fast and slow AI responses as a product quality gap. Teams building voice-first or chat-first interfaces for Indian markets should benchmark Flash against their target latency budgets before committing to an inference architecture.

The cost structure also makes Flash viable for high-throughput applications that would have been prohibitive at Gemini 3.0 pricing — document automation for financial services, multilingual support bots for e-commerce, and agent-assisted code review pipelines are all now within reach of early-stage product budgets.

The Bottom Line

Gemini 3.5 Flash is the most capable model Google has ever deployed at the low end of its API pricing. The combination of sub-$2 input pricing, one-million-token context, 289-tokens-per-second throughput, and benchmark scores that exceed the previous Pro tier creates a clear case for any team evaluating AI infrastructure in June 2026. Gemini 3.5 Pro, arriving this month, will extend that story at the frontier. For Indian product teams, the practical implication is direct: run your workload evals on Flash today, decide whether Pro is worth the wait, and move.