Google is bouncing back in the AI race. After falling behind OpenAI and Anthropic, the company has now released Gemini 3.8 Flash, claiming that it is its most advanced AI model when it comes to coding performance. Google says that Gemini 3.8 Flash can match advanced models like Claude Opus 5 and GPT-5.6 Sol but at a much lower cost.
Google CEO Sundar Pichai announced the release on X, stating that the model was the company’s “3rd Flash release in just 6 weeks.” Flash models are smaller models – usually trained on a few hundred billion parameters instead of trillions in case of larger models. They are also cheaper to run.
Pichai claimed that as per benchmarks, Gemini 3.8 Flash “outperforms most larger frontier models in autonomously solving complex engineering problems end to end, at a fraction of the cost.” He added that the model brought significant improvements compared to Gemini 3.7 Flash in tasks like software engineering and multi-step reasoning.
Alongside the standard Flash model, Google also launched Gemini 3.8 Flash Cyber, calling it its most capable cybersecurity model.
Gemini 3.8 Flash is cheaper to run
The company is offering an introductory price of 75 (roughly Rs 70) cents per million input tokens and $3.75 (roughly Rs 350) per million output tokens until December 31, 2026. Though Google says the model may use more tokens on harder tasks. After December 31, the rates will increase to $1.5 (roughly Rs 140) per million input tokens and $7.5 (roughly Rs 700) per million output tokens.
For context, Anthropic’s Claude Opus 5 would cost you $5 (roughly Rs 470) per million input tokens and $25 (roughly Rs 2350) per million output tokens. OpenAI’s GPT-5.6 Sol would set you back by $4 (roughly Rs 375) per million input tokens and $20 (roughly Rs 1890) per million output tokens.
The cost factor has become increasingly important in the AI race as companies rethink their AI bills. The lower costs may give Google an advantage in the enterprise market where coding remains one of the most important use cases. However, Chinese open-weight models like Kimi K3 have also gained traction recently for being even cheaper to run.
Gemini 3.8 Flash makes big gains
On HLE-Verified, Google said Gemini 3.8 Flash scored 54.9 per cent, which it said shows its ability to handle multi-step reasoning across STEM, humanities and professional fields. Its knowledge cut-off is March 2026 for some domains, while in others users may find its knowledge limited to January 2025.
Google said the gains come from a design choice that lets 3.8 Flash “work harder” by taking extra reasoning steps and calling tools iteratively on complex tasks. The company says this could raise overall usage costs even with unchanged token pricing, and said developers who want lower token overhead can use lower effort settings or continue with Gemini 3.7 Flash.
Artificial Analysis said Gemini 3.8 Flash is “the cheapest we’ve measured at this level of intelligence”, but added that total usage was up by about 40 per cent from 3.7 Flash because of higher output tokens and more turns on agentic evaluations.
Gemini 3.8 Flash is now live on the Gemini app for Google AI Pro and Ultra subscribers. It is also available to developers through Google Antigravity, AI Studio, and the Gemini API.
Google launches Gemini 3.8 Flash Cyber
Google also announced Gemini 3.8 Flash Cyber, which replaces 3.5 Flash Cyber for trusted defenders under the new Fairwind Program. The company said the model offers “frontier-level performance in autonomous vulnerability discovery” and focused on detection and patching rather than offensive capabilities.
Google said the Chrome Security team found that 3.8 Flash Cyber produced 2.6 times more correct patches to Chrome vulnerabilities than the best much larger commercial models. Google’s Cloud Vulnerability Research team said it used the model to find a critical foundational vulnerability in less than two hours, work that usually takes months.
This brings Google closer to rivals when it comes to cybersecurity uses for AI. Anthropic and OpenAI have made major progress with cyber-focused models in recent weeks. Recently, OpenAI claimed that its upcoming model Astra had crossed critical thresholds for cybersecurity.
The launch of Gemini 3.8 Flash and 3.8 Flash Cyber comes as Google looks to strengthen its position against OpenAI and Anthropic in coding-focused AI tools. The tech giant had fallen behind rivals and even reportedly delayed the release of its flagship model Gemini 3.5 Pro. At the same time, the company has seen some big leadership changes in its AI unit, with DeepMind’s Demis Hassabis stepping aside as CEO. Google is now said to be working on a new flagship model, Gemini 4.
