Google releases Gemini 3 Flash, promising improved intelligence and efficiency

Google releases Gemini 3 Flash, promising improved intelligence and efficiency

Google is ramping up its AI offerings with the introduction of Gemini 3 Flash, following the recent launch of the Gemini 3 Pro model. This enhanced version is set to be integrated into the Gemini app and search functionalities, with immediate access available for developers through the Gemini API, Vertex AI, AI Studio, and Antigravity. The new Gemini 3 Flash is not just faster; it promises improved capabilities over its predecessor. Google has shared various benchmark results showcasing its advancements, particularly in academic and reasoning assessments. For instance, it surpasses the previous 2.5 Flash model in tests like GPQA Diamond and MMMU Pro, even outperforming the 3 Pro model in some instances. One standout achievement for Gemini 3 Flash is its performance in Humanity’s Last Exam (HLE), where it has achieved a score of 33.7 percent without utilizing any tools, a significant improvement over previous models. This score is only slightly shy of the Gemini 3 Pro, emphasizing the new model’s advanced domain-specific expertise. Google has highlighted Gemini 3 Flash's coding capabilities, presenting it as a competitor to the Pro models that were previously considered the top choice for code generation. In the SWE-Bench Verified test, it has shown nearly a 20-point improvement over the 2.5 version. Furthermore, its performance in general-knowledge assessments has significantly improved, with a score of 68.7 percent in the Simple QA Verified test, a stark contrast to the 28.1 percent achieved by the last Flash model. In addition to its enhanced capabilities, Gemini 3 Flash is also more efficient than prior models. It retains the ability to generate interactive simulations and multimodal content, a hallmark of the Gemini 3 Pro. Operating at three times the speed of Gemini 2.5 Pro, it also offers a more cost-effective option for developers, with pricing set at $0.50 for one million input tokens and $3 for one million output tokens—though these rates have increased from the previous model’s pricing of $0.30 and $2.50, respectively. The Pro model remains at a higher cost, priced at $2 for one million input tokens and $12 for output. With these updates, Gemini 3 Flash positions itself as a formidable tool in Google’s AI arsenal, making significant strides in performance and accessibility for developers.

Sources : Ars Technica

Published On : Dec 17, 2025, 16:16

Cybersecurity
Hugging Face's CEO Demands Transparency Following OpenAI Cyber Incident

In a recent turn of events, OpenAI acknowledged a serious breach involving one of its models that affected the AI platfo...

TechCrunch | Jul 26, 2026, 17:10
Hugging Face's CEO Demands Transparency Following OpenAI Cyber Incident
Computing
Linux Creator Challenges Developers: Embrace AI or Fork Off!

In a recent discussion surrounding the use of AI in software development, Linus Torvalds, the founder of Linux, voiced h...

Business Insider | Jul 26, 2026, 13:10
Linux Creator Challenges Developers: Embrace AI or Fork Off!
Computing
Reclaiming Control: Librarians Host Workshops to Help People Navigate AI Tools

In a lively library setting in South Philadelphia, Charlie Bailey, a local librarian, humorously noted, "Everybody’s on ...

TechCrunch | Jul 25, 2026, 16:20
Reclaiming Control: Librarians Host Workshops to Help People Navigate AI Tools
AI
The Rise of AI Distillation: A Controversial Technique Sparks Debate in Tech and Government

In recent discussions, a once-obscure topic in artificial intelligence has surged to the forefront of debates among tech...

CNBC | Jul 25, 2026, 12:15
The Rise of AI Distillation: A Controversial Technique Sparks Debate in Tech and Government
Computing
Crisis Averted: Power Line Failure Highlights Urgent Need for Data Center Resilience

A power line failure near Washington, DC, recently showcased a significant challenge faced by the electrical grid due to...

TechCrunch | Jul 25, 2026, 13:50
Crisis Averted: Power Line Failure Highlights Urgent Need for Data Center Resilience
View All News