Google Launches Gemini 3.7 Flash as AI Coding and Business Agents Become More Powerful

Google has released **Gemini 3.7 Flash**, a new AI model designed specifically for coding, software development and automated workflows.

Announced on 13 August 2026, the model is being positioned as Google's latest high-speed "workhorse" for developers and businesses building AI agents capable of planning tasks, using software tools and completing multi-step processes with less human intervention.

The release comes only three weeks after Gemini 3.6 Flash, at a time when Google is under pressure to keep pace with OpenAI and Anthropic in AI coding and agentic software.

Gemini 3.7 Flash is built for agents

Google is not presenting Gemini 3.7 Flash simply as a chatbot model. Its main target is agentic workflows — where a system is given a goal and then uses tools, software and multiple steps to accomplish it.

Rather than responding to one prompt at a time, the model can plan, execute actions, evaluate results and continue working. Google says it has been improved for these longer workflows, including better instruction following, planning and tool use.

That matters because the industry is shifting from systems that generate answers toward systems that perform work.

Software development is a core focus

Google says Gemini 3.7 Flash delivers substantial improvements in debugging, issue resolution and production-code generation compared with Gemini 3.6 Flash, reporting first-pass code-generation results of:

• 43.6% on FrontierCode 1.1 Main, compared with 34.4% for Gemini 3.6 Flash
• 65.3% on DeepSWE v1.1, compared with 49.0% for Gemini 3.6 Flash
• 1588 Elo on WebDev Arena for web development, compared with 1538 previously
• 30.4% on AutomationBench for enterprise workflow automation, compared with 17.0%

These are Google's own reported benchmarks, so they should be read as vendor comparisons rather than independent proof that the model will outperform competitors on every real-world task.

Google is making it cheaper to run AI agents

One of the most notable parts of the launch is pricing. Google is offering Gemini 3.7 Flash at an introductory rate of $0.75 per 1 million input tokens and $3.75 per 1 million output tokens through the end of 2026 — half the original cost of Gemini 3.6 Flash.

Lower inference costs matter because agents make many model calls to finish a single task. A chatbot might generate one response; an agent could perform dozens of reasoning steps, call APIs, inspect files, run code and revise its output. The cheaper each interaction becomes, the more viable it is to deploy agents at scale.

Businesses look like the main target

Consider an AI agent that receives a customer enquiry and then reads the request, checks the CRM, verifies product availability, generates a quotation, updates the customer record and sends a follow-up message. That is considerably more powerful than a conventional chatbot.

Gemini 3.7 Flash is also being deployed inside Gemini Spark, Google's subscription AI agent service, which can operate under user direction across applications — consolidating files, drafting emails and updating status documents. That gives Google control of the model, the agent and the workspace ecosystem, a potentially major advantage in enterprise AI.

The opportunity for Kenyan businesses

Many Kenyan SMEs still depend heavily on manual processes involving email, WhatsApp messages, spreadsheets, phone calls and disconnected business applications. Agentic AI could connect those processes end to end:

• Customer enquiry → AI agent → CRM → stock check → quotation → WhatsApp follow-up
• Job application → document analysis → candidate database → shortlist → interview scheduling
• Invoice received → AI extraction → accounting system → verification → approval workflow

The value is not in simply having an AI assistant. It is in connecting AI to the systems that already run the business.

There are still important limitations

Marketing around AI agents can make these systems sound more autonomous than they currently are. Businesses still need to think carefully about accuracy, security, permissions, privacy, monitoring, failure handling and human oversight.

Agentic AI becomes far more powerful when given access to business systems — but that same access increases the consequences of mistakes.

Our take

Gemini 3.7 Flash is important not because Google shipped another model, but because Google is increasingly optimising its AI for software engineering and autonomous workflows. Better coding, stronger tool use and lower costs together make agents materially more practical.

The metric to watch will not be benchmark scores. It will be whether companies can reliably hand agents real work and get a useful result without supervising every step. AI is moving from generating software to operating software — and that may be one of the most important shifts in business automation of the next few years.