谷歌AI发布Gemini 3.7 Flash：编码与智能体模型，输入每百万tokens仅0.75美元

Google AI Just Released Gemini 3.7 Flash: A Coding and Agent Model at $0.75/1M Input Tokens

Google has released Gemini 3.7 Flash, the newest model in its Flash tier, three weeks after Gemini 3.6 Flash. The model card describes it as a refinement of 3.6 Flash with algorithmic improvements to the core reasoning foundation — not a new pretraining run. It accepts text, images, audio, and video across a 1M-token context window, returns up to 64K output tokens, and supports customizable thinking configurations that trade quality against cost and latency. The knowledge cutoff stays at March 2026. The gains concentrate in three places: software engineering, document-heavy knowledge work, and web development. The sharper argument is price. Gemini 3.7 Flash ships at $0.75 per 1M input tokens and $3.75 per 1M output tokens — half the original 3.6 Flash list rate, and roughly a third the blended cost of Claude Sonnet 5 or GPT-5.6 Terra.

Yes, API and enterprise only. There are no open weights. Access runs through hosted surfaces: the Gemini API and Google AI Studio, Google Antigravity, Android Studio, the Gemini Enterprise Agent Platform, and the Gemini Enterprise app. Consumers reach it through Gemini Spark on Google AI Pro and Ultra plans.

Company fit: Startups and mid-market teams gain the most, because the introductory price makes always-on agents affordable without a Pro-tier budget. Regulated enterprises get a governed path through Gemini Enterprise. Teams with data-residency or air-gap requirements are excluded — there is nothing to self-host.

Industries: Googles own eval set points at legal, financial services, biosciences, and enterprise operations. The Harvey LAB-AA, GDP.pdf, and AutomationBench results are the tells.

Applications: Long-running coding agents, document-heavy back-office automation, UI generation from screenshots or design systems, and PDF-to-structured-data pipelines.

On FrontierCode 1.1 Main, which measures production code quality, Gemini 3.7 Flash scores 43.6% against 34.4% for 3.6 Flash. On DeepSWE v1.1, a long-horizon software engineering eval, it reaches 65.3%. On WebDev Arena it posts an Elo of 1588 versus 1538, the top score in Googles comparison table.

Document and workflow results move further. GDP.pdf, an expert PDF comprehension eval, goes from 22.0% to 34.0%. AutomationBench, a private enterprise workflow set, goes from 17.0% to 30.4% — ahead of both Claude Sonnet 5 at 10.7% and GPT-5.6 Terra at 23.6%. Long-context retrieval on GDM-MRCR v2 at 128k reaches 97.0%.

GPT-5.6 Terra is ahead on DeepSWE (69.6%), Terminal-bench 2.1 (87.4%), Terminal-bench 3.0 (20.8%), and OSWorld-2.0 (50.2%). On GDPval-AA v2 knowledge work, 3.7 Flash scores 1525 Elo against 1598 for Sonnet 5 and 1628 for Muse Spark 1.2. CharXiv Reasoning is a regression: 84.5% without tools, down from 85.2% for 3.6 Flash. On the Artificial Analysis Intelligence Index, 3.7 Flash scores 56, against 57 for both GPT-5.6 Terra and Muse Spark 1.2.

Gemini 3.7 Flash lists at $0.75 per 1M input tokens and $3.75 per 1M output tokens. That rate is introductory and expires December 31, 2026; from January 1, 2027 it becomes $1.50 and $7.50. In Google's own table, Claude Sonnet 5 sits at $2.00/$10.00 and GPT-5.6 Terra at $2.00/$12.00.

At an 80/20 input-output mix, that is a blended $1.35 per 1M tokens today against $3.60 for Sonnet 5 and $4.00 for GPT-5.6 Terra. For teams running agents at volume, the intelligence-per-dollar gap is the reason to evaluate, not the individual eval wins.

Gemini 3.7 Flash is a refinement of 3.6 Flash, not a new base model, shipped just three weeks later.

Coding gains are real: FrontierCode 43.6% vs 34.4%, DeepSWE 65.3% vs 48.6%, WebDev Arena 1588 Elo.

Price is the strongest claim — $0.75/$3.75 per 1M until December 31, 2026, then it doubles.

GPT-5.6 Terra still leads on terminal and computer-use agents; CharXiv is a small regression.

API and enterprise only. No open weights, so no self-hosting or air-gapped deployment.

Check out the Technical Details. Also, feel free to follow us on Twitter and don’t forget to join our 150k+ML SubReddit and Subscribe to our Newsletter. Wait! are you on telegram? now you can join us on telegram as well.

Need to partner with us for promoting your GitHub Repo OR Hugging Face Page OR Product Release OR Webinar etc.? Connect with us

Asif Razzaq is the CEO of Marktechpost Media Inc.. As a visionary entrepreneur and engineer, Asif is committed to harnessing the potential of Artificial Intelligence for social good. His most recent endeavor is the launch of an Artificial Intelligence Media Platform, Marktechpost, which stands out for its in-depth coverage of machine learning and deep learning news that is both technically sound and easily understandable by a wide audience. The platform boasts of over 2 million monthly views, illustrating its popularity among audiences.

Liquid AI Releases LFM2.5-VL-3B: A 3B Vision-Language Model That Reads Screens, Grounds Objects, and Calls Tools On-Device

Dyna Robotics Introduces Dyna-2: A World-Action Model Pre-Trained on 1 Million Hours of Human Video

NVIDIA AI Releases Nemotron 3.5 Lightning: A 30B Open MoE with 3B Active Parameters, and NeMo Switchyard Model Router

The Video Production Stack Now Fits on One Desk: LTX-2.5 Launches as NVIDIA-Accelerated Open Weights World Model

Previous articleLiquid AI Releases LFM2.5-VL-3B: A 3B Vision-Language Model That Reads Screens, Grounds Objects, and Calls Tools On-Device