Gemini 3.8 Flash Starts Testing Google, Focuses on Coding and Reduces AI Hallucinations
Google is reportedly starting to test Gemini 3.8 Flash internally, just about two weeks after the company launched Gemini 3.7 Flash on August 13, 2026.
A Business Insider report said that a number of Google employees have used the early version of Gemini 3.8 Flash through Jetski, the company's internal code development platform. The technicians tested the unannounced model in various programming tasks before it was possibly opened through an API for external users.
Google has not released official performance benchmarks or card models for the Gemini 3.8 Flash. However, early reports indicate that the new model is aimed at improving capabilities compared to the Gemini 3.7 Flash.
The development of Gemini 3.8 Flash is said to focus on a number of problems commonly found in lightweight AI models. Google is reportedly improving the ability to run multi-step agent workflows, speed up code generation, and reduce hallucinations in long conversations.
Another clue about the development of the model comes from reports from whistleblowers. One of them said that Google had run A/B tests with a number of partners to improve the model's ability to run various tools.
This accelerated development took place after Google launched Gemini 3.6 Flash on July 21, 2026, which was followed by Gemini 3.7 Flash about three weeks later. Logan Kilpatrick, Google AI Studio product lead, said the improvement was the result of algorithm refinements and not a full retraining of the model.
Gemini 3.7 Flash was launched by Google on August 13, 2026. In its first week, Google CEO Sundar Pichai called it Google's fastest growing model. The model was then used to support a number of features in Search and the Gemini application.
Gemini 3.7 Flash also reportedly recorded high performance in the AnalystAgent benchmark from Artificial Analysis for spreadsheet processing tasks.
Reports on the Gemini 3.8 Flash came as Google was said to be speeding up the development of the Flash model line. The Gemini 3.5 Pro was previously reported to have been canceled or delayed, while the Gemini 4.0 is said to still need time before being launched.
If the report proves true and Google opens the public preview version of Gemini 3.8 Flash in September, the company could potentially release three updates to the Gemini Flash model in less than two months.
The English, Chinese, Japanese, Arabic, and French versions are automatically generated by the AI. So there may still be inaccuracies in translating, please always see Indonesian as our main language. (system supported by DigitalSiber.id)
Related Stories
AI News
Senior News Tackles ‘Artificial Intelligence’
43 minutes ago
AI News
We tested how AI chatbots would handle foreign propaganda. They did surprisingly well
43 minutes ago
AI News
China promotes coordinated development of data industry, artificial intelligence
1 hour ago
AI News
Losing face? Licensing and governing AI content in China
1 hour ago
AI News
Lightspeed Commerce Appoints Bhawna Singh as Chief Technology Officer; Ships Wave of AI, Payments and Wholesale Innovations
3 hours ago
AI News
Microsoft Tokenmaxxing King Spent $28,000 on AI in a Single Month as Company Begs Employees to Tone Down the Ludicrous Costs
3 hours ago
AI News
CN train derails in Truro, blocking key crossing; no injuries reported
6 hours ago
AI News
OpenAI Cancels Cursor Partnership Citing Distrust of Elon Musk
6 hours ago