Google Launches Gemini 3.5 Flash, Claims 4x Faster Than Rivals

Google has launched Gemini 3.5 Flash, a new AI model focused on speed and cost-efficiency, now available across Google Cloud and consumer services. The company claims the model is approximately four times faster and often less than half the cost of rival models, according to its I/O 2026 blog post (Google Blog) and a related Cloud recap (Google Cloud Blog). Described as its strongest agentic and coding model yet, Gemini 3.5 Flash is already the default engine in the Gemini app and Search's AI Mode (TechCrunch).

Early enterprise pilots

Initial enterprise pilots demonstrate how Gemini 3.5 Flash's speed and cost benefits can translate to business value. Google states these partners are already "seeing meaningful impact" as the model handles planning, tool execution, and error recovery in extended, complex workflows. Key adopters include:

• Shopify: Refining merchant growth forecasts at scale using parallel subagents.
• Macquarie Bank: Testing low-latency agents to analyze 100-page onboarding documents.
• Salesforce: Integrating Flash into its Agentforce platform for multi-turn enterprise automation.
• Ramp: Pairing multimodal OCR with historical data for more reliable invoice extraction.
• Xero: Automating supplier discovery and tax form preparation over multi-week horizons.
• Databricks: Monitoring large datasets to retrieve context and propose real-time fixes in data pipelines.

Gemini 3.5 Flash is Google's latest AI model, optimized for high speed and low cost. It is designed for rapid, high-volume tasks like data extraction, summarization, and agentic workflows. Google claims it runs four times faster than comparable models while costing less than half the price.

Benchmark context

Gemini 3.5 Flash's performance is highlighted by strong scores on key agentic benchmarks, which measure a model's ability to complete multi-step jobs. Flash scored 76.2% on Terminal-Bench 2.1, a benchmark that tests an agent's ability to navigate command-line environments and solve technical problems. This score is notable, as previous models rarely surpassed 70%, though experts caution that real-world success rates can be lower than lab results. On MCP Atlas, a test for dynamic tool use, Flash achieved an 83.6%, surpassing Google's prior models and competing with rival scores. Additional results position Flash as a competitive model for specialized reasoning tasks.

Price-performance claims and roll-out channels

Google claims Gemini 3.5 Flash is priced at "less than half the cost of comparable models," aiming to establish a new price-performance standard. The model is immediately available to developers via Google Cloud platforms like Gemini Enterprise Agent Platform and AI Studio. For consumers, it is accessible in Gemini Spark and Search's AI Mode. This launch precedes the upcoming Gemini 3.5 Pro, scheduled for release next month. Google positions Flash for rapid, high-volume tasks, while Pro will be geared toward tasks requiring greater reasoning depth.

What to watch

The industry will be closely monitoring whether Flash's impressive benchmarks translate into tangible productivity gains. Key indicators to watch include:

1. Workflow Completion: How consistently the model completes multi-week tasks compared to human performance.
2. Cost-Effectiveness: The actual cost per automated task, particularly in the initial finance and e-commerce pilots.
3. Reliability: The model's stability and consistency across repeated runs, especially for coding and data operations.

Google is also expected to release further studies on agent alignment, which will measure not just task success but also adherence to instructions in real-world environments.

---

What is Gemini 3.5 Flash and why is Google calling it a breakthrough?

Google's Gemini 3.5 Flash is a new AI model designed for high-speed performance and cost-efficiency, announced at I/O 2026. It is called a breakthrough due to its claims of being four times faster and less than half the price of comparable models. Its capabilities are supported by high scores (76.2% on Terminal-Bench 2.1 and 83.6% on MCP Atlas) on benchmarks that test complex, agentic task completion.

Which enterprises are already running Gemini 3.5 Flash in production?

Leading enterprises including Shopify, Macquarie Bank, Salesforce, Ramp, Xero, and Databricks are early adopters. They are using Flash for complex, long-running agentic tasks, such as Shopify refining growth forecasts, Macquarie Bank performing compliance checks on large documents, and Databricks monitoring and fixing data pipelines.

How do the new benchmarks Terminal-Bench 2.1 and MCP Atlas work?

These benchmarks evaluate an AI model's ability to act as an "agent" that completes a job, not just answer a question. Terminal-Bench 2.1 assesses performance in realistic command-line environments on tasks like DevOps and debugging. MCP Atlas tests the model's skill in using multiple tools (like APIs and databases) over hundreds of steps to achieve a goal.

What are the exact cost and speed claims versus competitors?

Google claims Gemini 3.5 Flash is approximately four times faster and priced at less than half of comparable frontier models from competitors like OpenAI and Anthropic. While independent verification is ongoing, initial benchmark data suggests its performance is within a few percentage points of top-tier models on agentic tasks, but at a significantly lower price point.

When will Gemini 3.5 Pro ship and how will it differ?

Gemini 3.5 Pro is scheduled to launch in June 2026 as a more powerful counterpart to Flash. While Flash is optimized for speed, the Pro version will focus on enhanced reasoning depth and capability for more complex tasks. It is expected to surpass the already strong performance metrics set by the Flash model.