Urgent.News

What's breaking now, across thousands of outlets.

AI

With Gemini 3.8 Flash, Google reminds everyone it's still in the race

AI model scores well, runs fast, and doesn't cost too much (yet)

With Gemini 3.8 Flash, Google reminds everyone it's still in the race

Google unveiled Gemini 3.8 Flash on Wednesday in a bid to reestablish its standing as a leading model maker. The company's determination to compete in the frontier model race has been questioned since early August when DeepMind CEO Demis Hassabis stepped down and Koray Kavukcuoglu assumed a less prominent leadership role. Google's assurances failed to alleviate concerns following the delayed release of Gemini 3.5 Pro in June and the underwhelming benchmark results of Gemini 3.5 Flash, which, while quick and economical, have been surpassed by open-weight models from Chinese AI firms in recent months.

However, Google has accelerated its development efforts, with Gemini 3.8 Flash being its fourth Flash model in four months. This iteration has shown significant progress over its predecessor, often matching the performance of more costly frontier models at a fraction of the expense. On DeepSWE v1.1, a test for long-horizon software engineering, 3.8 Flash surpassed most larger frontier models, only at a lower cost.

The model scored 59 on the Artificial Analysis Intelligence Index, matching the intelligence level of GPT-5.6 Sol and Grok 4.6. When priced at $0.58 per Intelligence Index task, Gemini 3.8 Flash is the most affordable model at this level of intelligence, 40% more expensive than its predecessor due to increased output tokens and processing.

It delivers better results in enterprise knowledge work benchmarks, such as Vals Finance Agent V2, Harvey's Legal Agent Benchmark, and HLE-Verified. Gemini 3.8 Flash is accessible to developers through Google Antigravity, the Gemini API in AI Studio and Android Studio, and Stitch's interface design service. Enterprises can utilize Gemini Enterprise, while consumers with Google AI Pro and Ultra subscriptions can access it via the Gemini app, AI Mode in Google Search, and Gemini within Google Sheets.

A program called Fairwind offers access to Gemini 3.8 Flash Cyber, a version of the model tailored for software vulnerability hunting and remediation, to governments, critical infrastructure organizations, and software maintainers.

Written by urgent.news from The Register's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.

This story

This is one outlet's version. Read the fullest account.

Read the original at theregister.com →

More in AI

Why LLM Memory in Production Fails Silently

Your agent's memory layer will not throw. It returns three plausible looking chunks, the model answers confidently from them, and nobody notices for a week.

  • LLM memory systems fail silently in production.
  • Retrieval quality degrades over time as corpus grows.
  • Verification hooks can detect failures by checking quality metrics.

More from Wednesday 2 September →