Google launched Gemini 3.7 Flash just three weeks after the 3.6 release, with a DeepSWE score of 65.3% and a launch price set at half that of its predecessor.

Key takeaways:

  • Gemini 3.7 Flash boosts its DeepSWE v1.1 score to 65.3% and reaches 1,588 Elo in web development.

  • The promotional rate is $0.75 per million input tokens and $3.75 per million output tokens through 2026.

  • The model is currently being deployed on Spark and is available across Google’s AI offerings for developers and businesses.

Launch of Gemini 3.7

Google announced Gemini 3.7 Flash on August 13, barely three weeks after Gemini 3.6 Flash. The group explains that developer feedback and algorithmic adjustments drove this iteration, with a priority focus on software engineering, web development, and knowledge tasks. The release pace is unusually sustained.

The gains are most visible in programming. DeepSWE v1.1 jumps to 65.3%, versus about 49% for 3.6 Flash, while FrontierCode 1.1 Main rises from 34.4% to 43.6%. Web development follows the same trend, with an Elo WebDev Arena up from 1,538 to 1,588.

Google says the model “generates more functional interfaces and complete applications in fewer requests.” Performance on knowledge tasks also improves. The GDP.pdf benchmark rises from 22.0% to 34.0%, while AutomationBench climbs from 17.0% to 30.4%. Google says it has strengthened its safeguards, especially against CBRN use cases and cyber offensives.

Also read: Anthropic targets a $6B Decart deal while preparing for its stock market entry

AI performance

External tests confirm these advances. Niko Grupen, head of applied research at Harvey, highlights an improvement of 2.6 points in the overall success rate compared to 3.6 Flash on the Legal Agent Bench benchmark. Kyrill Hux, cofounder of Nunu.ai, meanwhile estimates that internal measures place 3.7 Flash at the level of mid-sized frontier models, for a cost about cut in half.

The price drop reinforces this argument, since high-volume agents can generate substantial costs through successive loops of planning, tool calls, and follow-ups. Google says 3.7 Flash follows instructions more strictly and handles blockers better, which could reduce the need for human intervention in engineering and management workflows. The model is now available generally.

Gemini 3.6 Flash was launched on July 21 at $1.50 per million input tokens and $7.50 per million output tokens.

Three weeks later, 3.7 Flash reaches $0.75 and $3.75 respectively, until December 31, while improving its benchmark scores and being deployed on Spark for AI Pro and Ultra subscribers. This rapid sequence extends Google’s accelerated update cycle for its Flash models.

Coming up: Trading volume for Virtuals Protocol explodes by 180%, and the bulls aim for $0.68