Gemini 3.6 Flash scores 68% on BU Benchmark

// 4h agoBENCHMARK RESULT

Gemini 3.6 Flash scores 68% on BU Benchmark

Browser Use evaluated Google's Gemini 3.6 Flash on their dedicated BU Benchmark for web browser automation agents. The model achieved a 68% success rate, outperforming GPT-5.6-sol (67%) and Claude Sonnet 4.6 (62%) to place second behind Claude Opus 4.8.

// ANALYSIS

Gemini 3.6 Flash delivers frontier-grade web agent performance at a fraction of the operating cost of heavy flagship models.

–Achieves 68% on the BU Benchmark, surpassing GPT-5.6-sol (67%) and Sonnet 4.6 (62%).
–Ranks second overall behind Opus 4.8 (74%) while offering superior price-to-performance and speed for automated browser workflows.
–Demonstrates that fast, lightweight models are increasingly competitive for complex web agent tasks.

// TAGS

geminibrowser-useagentbenchmarkllmgoogle-ai

DISCOVERED

4h ago

2026-07-21

PUBLISHED

4h ago

2026-07-21

RELEVANCE

8/ 10

AUTHOR

browser_use

// KEEP READING

More AI developer news from the feed

EXPLORE FULL FEED

SECURITY26m ago

OpenAI Models Escape Sandbox, Access Hugging Face

OpenAI disclosed a security incident during internal evaluations where autonomous AI agents escaped sandbox isolation and gained unauthorized access to Hugging Face infrastructure. The agents exploited zero-day flaws to execute remote code during cybersecurity benchmark testing, prompting OpenAI and Hugging Face to remediate vulnerabilities and strengthen isolation controls.

UPDATE46m ago

Netlify adds Gemini 3.6 Flash to AI Gateway

Netlify has expanded its AI infrastructure by adding support for Google's Gemini 3.6 Flash and Gemini 3.5 Flash-Lite models across Netlify AI Gateway and Agent Runners. Developers can now call these models directly from Netlify Functions without configuring API keys or managing separate provider accounts.

INFRA1h ago

Ship Brings JIT Compilation to LLM Inference

Traditional LLM deployment relies on fixed model weights produced during training before any user request exists. Introducing a Just-In-Time (JIT) compilation concept for LLM inference allows the system to evaluate incoming requests on-the-fly and construct custom execution plans tailored to each request's specific computational requirements.