Google’s Gemini 3.7 Flash targets coding and agents with a 50% introductory price cut

What happened

I'll pull the VentureBeat piece and Google's own notes so the paragraphs stay factual and specific, then write the body in plain prose.I'll draft the body from the VentureBeat piece and Google's own numbers, then count words so it stays in the 600–900 range.Slightly over the cap — I'll trim a few sentences so it lands inside 600–900 words.Google is rolling out Gemini 3.7 Flash, a new version of its workhorse AI model that puts coding, agentic workflows, and knowledge work at the center of the upgrade, while temporarily cutting API prices in half. The release arrives just three weeks after Gemini 3.6 Flash. Through the end of 2026, Gemini 3.7 Flash costs $0.75 per million input tokens and $3.75 per million output tokens. Starting January 1, 2027, those rates rise to $1.50 and $7.50. Google is applying the same introductory rate to Gemini 3.6 Flash. Context caching is $0.075 per million tokens during the promotional period and $0.15 afterward. The model is generally available through the Gemini API in Google AI Studio and Android Studio, Google Antigravity, Gemini Enterprise Agent Platform, and the Gemini Enterprise app. Google AI Pro and Ultra subscribers get it in Gemini Spark.

Google describes Gemini 3.7 Flash as its most intelligent workhorse model yet for coding and agents. It says the model is better at adapting when it hits roadblocks, clarifying intent, and following instructions with greater fidelity. The stated mechanic is more diligent thinking: more effort on multi-step planning and tool calls, aimed at fewer retries and less manual supervision. That is a change from Gemini 3.6 Flash, which Google documented as reducing reasoning steps, conversational turns, and tool calls to limit execution-loop spiraling. Gemini 3.7 Flash supports a 1 million token context window, 64,000 maximum output tokens, and thinking levels of low, medium, and high, with medium as the default. High thinking extends thoughts and function calls for harder coding and agent tasks and spends more tokens. The model is now the default behind the Antigravity agent.

The technical detail

For engineers, the useful claim is operational. A coding agent that makes fewer unnecessary edits, recovers from errors, and executes a multi-step plan without looping can cut human interventions per ticket. The same property matters for a knowledge-work agent that must read a long report, pick a tool, update another system, and produce a document for review. Google reports FrontierCode 1.1 Main at 43.6 percent, up from 34.4 percent on Gemini 3.6 Flash. On DeepSWE v1.1, a long-horizon software engineering eval, the score is 65.3 percent versus 49.0 percent for the predecessor. On Code Arena, Gemini 3.7 Flash scores an Elo of 1588 against 1538 for 3.6 Flash. AutomationBench rises from 17.0 percent to 30.4 percent. GDP.PDF, a complex PDF comprehension test, rises from 22.0 percent to 34.0 percent. Browser Use said its Gemini 3.7 Flash agent was 35 percent cheaper than 3.6 Flash, with an 8-point increase in prompt-cache hit rate and fewer tool errors.

Why it matters for builders

Google's own comparisons do not show Gemini 3.7 Flash displacing every higher-priced rival. GPT-5.6 Terra still leads DeepSWE v1.1 at 69.6 percent and Terminal-bench 2.1 at 87.4 percent against 85.8 percent for Gemini 3.7 Flash. Claude Sonnet 5 leads Agent's Last Exam multimodal desktop and operating-system tasks at 33.3 percent versus 26.3 percent. On FrontierCode, 43.6 percent for Gemini 3.7 Flash narrowly exceeds 42.7 percent for Claude Sonnet 5 and 41.3 percent for GPT-5.6 Terra. Google lists Claude Sonnet 5 at $2 and $10 per million input and output tokens and GPT-5.6 Terra at $2 and $12. Artificial Analysis places Claude Opus 5 at 63 on its Intelligence Index and Gemini 3.7 Flash at 56, up from 52 for Gemini 3.6 Flash. The competitive offer is a cheaper workhorse that is close on production coding, web layouts, and document-heavy agents, not a sweep of every frontier eval.

Market and competitive context

Treat the half-price window as a measurement period, not a permanent unit cost. Introductory pricing expires December 31, 2026. After that, both Gemini 3.7 Flash and Gemini 3.6 Flash return to $1.50 and $7.50. Score cost per successfully completed task on your own repositories, tool schemas, and PDF-heavy workflows, and do it at medium and high thinking levels. High thinking can erase the promotional discount if extra reasoning tokens do not close the loop. Watch retry rate, failed agent loops, and cache hit rate as closely as pass rate. Also watch whether Google keeps shipping Flash upgrades on a three-week cadence. Gemini 3.5 Pro still has no public date. The latest released general-purpose Pro model remains Gemini 3.1 Pro from February, and Google has described Gemini 4 as its most ambitious training run.

What to watch next

The release sits next to a delayed flagship and an organizational reset. Demis Hassabis has given up day-to-day control of DeepMind to become its chair and Alphabet chief scientist. Former DeepMind CTO Koray Kavukcuoglu now runs the unit as a senior vice president reporting to Sundar Pichai. Jeff Dean, Oriol Vinyals, Quoc Le, and Sanjay Ghemawat left to form Discovery Loop. Noam Shazeer moved to OpenAI and John Jumper to Anthropic. Reuters reported that Gemini 3.5 Pro missed its original target after falling short of internal goals, particularly in coding. SemiAnalysis has argued Google is prioritizing cloud infrastructure sales to AI companies and claimed 3.5 Pro was effectively canceled. Google has not confirmed that and still describes the model as delayed. Gemini 3.7 Flash ships with updated safeguards covering chemical, biological, radiological, and nuclear risks and cyber-offense misuse. The open question for engineering orgs is whether a half-price Flash model that still trails on some terminal and desktop-agent evals can replace a Claude Sonnet 5 or GPT-5.6 Terra default, or whether it only wins the high-volume inner loop until January 1, 2027.