Monday, 3 August 2026 Sign in

Gemini 3 Pro

Model ● Covered in 10 stories Compare ⇄ + Follow

Gemini 3 Pro is Google's AI model released in late 2025 for complex reasoning and multimodal tasks across consumer, developer, and enterprise applications. The model has been evaluated across multiple benchmarks, achieving a FACTS Score of 68.8% for factuality, 54.2% on Terminal-Bench 2.0 for tool use, and scoring 1501 Elo on the LMArena Leaderboard, while showing limited performance on specialized tasks like writing optimized multi-GPU kernels where it solved only 28 of 87 real-world problems in zero-shot evaluation.

Updated 3 August 2026

Specifications

No specifications recorded yet.

Latest developments

Cisco Foundation AI Releases Antares: 350M and 1B Open-Weight Models That Localize Known Vulnerabilities Inside Real Codebases MarkTechPost · 1 week ago · 24

ParallelKernelBench: Frontier LLMs can't write fast multi-GPU kernels (yet) Together AI · 1 month ago · 12

Protecting people from harmful manipulation Google DeepMind · 4 months ago · 7

Gemini 3.1 Pro: A smarter model for your most complex tasks Google DeepMind · 5 months ago · 1

A Guide to Which AI to Use in the Agentic Era One Useful Thing · 5 months ago · 22

Google's year in review: 8 areas with research breakthroughs in 2025 Google DeepMind · 7 months ago · 35

FACTS Benchmark Suite: Systematically evaluating the factuality of large language models Google DeepMind · 7 months ago · 33

Introducing Nano Banana Pro Google DeepMind · 8 months ago · 6

Timeline

Month Quarter Year

Q3 2026

22 Jul Cisco Foundation AI Releases Antares: 350M and 1B Open-Weight Models That Localize Known Vulnerabilities Inside Real Codebases

Q2 2026

23 Jun ParallelKernelBench: Frontier LLMs can't write fast multi-GPU kernels (yet)

Q1 2026

Q4 2025

Google DeepMind releases Nano Banana Pro, an image generation and editing model based on Gemini 3 Pro Model release

Google releases Gemini 3 AI model with multimodal capabilities and agentic features Model release

Relationships

Products & technology

Google develops this model · 6 sources
Nano Banana Pro derived from this model · 1 source
Gemini 3.1 Pro derived from this model · 1 source
Google DeepMind develops this model · 1 source

The daily briefing

Every AI story that matters,
in your inbox by 8am.

TLDRocket reads 60+ sources, removes duplicate coverage, and summarises the day in two minutes. Follow companies and topics for alerts, or get the briefing in Slack. Free, no spam, unsubscribe anytime.

TLDRocket 3 August 2026

Open-source AI models are consolidating their foothold as the dominant deployment pattern, even as frontier capabilities remain concentrated at a handful of well-funded labs. Interconnects launched the Artifacts Hub and Adoption Dashboard to track 792 open-weight models released in two years, mapping adoption by geography and revealing how deeply DeepSeek and Chinese models have penetrated deployment pipelines globally. Meanwhile, DeepSeek itself proved that scale isn't everything: its V4-Flash model, a leaner 284-billion-parameter variant, outperformed its own flagship on agent benchmarks through careful post-training. The MIT-licensed release lets enterprises avoid vendor lock-in entirely, directly controlling deployment and retraining at lower infrastructure cost.

Read the full briefing →