Gemini 3.7 Flash upgrades Spark with better planning and lower pricing

Google released Gemini 3.7 Flash, the latest workhorse model in its Gemini 3 family, and immediately put it to work powering the Spark agent. The update brings meaningful improvements in planning and tool use for multi-step workflows, and introductory pricing through December 31 makes it the cheapest option among comparable models from Anthropic and OpenAI.

Better planning, fewer retries for agent workflows

Gemini 3.7 Flash is designed to spend more effort on planning and tool calls, adapt better when it hits roadblocks, and need fewer retries and less manual intervention. Google says the model is a direct result of developer feedback and algorithmic innovations. For Spark, that means improved accuracy and output quality for complex, multi-skill workflows across Gmail, Drive, Docs, Calendar, Keep, and Tasks.

Spark can already continue working after a laptop is closed or a phone is locked. The Chrome extension takes this further: Spark can use accounts you are already signed into and passwords saved in the browser to navigate websites, compare flight options, schedule apartment viewings, and begin bookings. Sensitive actions like payments still require user approval.

Benchmarks show gains, but real-world reliability is the test

Google published benchmark results that position Gemini 3.7 Flash ahead of Claude Sonnet 5 and GPT-5.6 Terra on several agent and document tasks. On AutomationBench, Gemini 3.7 Flash scored 30.4%, up from 17% for Gemini 3.6 Flash, and ahead of GPT-5.6 Terra at 23.6% and Claude Sonnet 5 at 10.7%. On GDP.pdf, a document-heavy benchmark, it scored 34%, compared to 22% for Gemini 3.6 Flash, 28% for Claude Sonnet 5, and 24.7% for GPT-5.6 Terra. Coding benchmarks also improved: FrontierCode at 43.6% edged out Claude Sonnet 5 at 42.7% and GPT-5.6 Terra at 41.3%.

These numbers are vendor-reported and may not translate directly to every production workflow. The real test for builders will be whether the improved planning and reduced retries hold up in their specific agent pipelines.

Pricing that changes the cost equation

Introductory pricing through December 31 sets Gemini 3.7 Flash at $0.75 per million input tokens and $3.75 per million output tokens. That is roughly 60% cheaper than Claude Sonnet 5 at $2 and $10, and GPT-5.6 Terra at $2 and $12. For teams running long agent sessions or high-volume document processing, this pricing could significantly reduce operating costs. The catch: it is time-limited, and post-offer pricing has not been announced.

What's missing: Gemini 3.5 Pro remains delayed

Google previewed Gemini 3.5 Pro at I/O in May, but it has not shipped after missing its expected June rollout. The company continues to trail Anthropic and OpenAI at the frontier. For now, Gemini 3.7 Flash is the most capable workhorse model available, but builders waiting for a larger reasoning model will have to keep waiting.