DeepSeek V4 Flash is no longer text-only 👀 In our internal benchmarks, it delivered significantly better price-performance than others in its class. We’ve added vision for screen-level understanding, capabilities needed for WebBrain Available on HF: huggingface.co/webbrain-one... #deepseek #ai

webbrain-one/DeepSeek-V4-Flash-0731-Vision-NVFP4 · Hugging Face

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

huggingface.co

We tested 13 OpenRouter models across 1,300 browser-planning calls. DeepSeek V4 Flash led consensus, Gemini 3.6 Flash delivered 100/100 valid calls, and Kimi K3 agreed with Sonnet 5 on 86% of tool choices. Full benchmark: www.webbrain.one/blog/america...

Two AI gaps are narrowing: our thirteen-model planner benchmark

Thirteen OpenRouter models, 1,300 comparable calls, no reference-model judge: consensus quality, latency, modality, architecture, parameter count, and real test cost.

webbrain.one

🚀 WebBrain 26.0.2 is out — built with contributions from 6+ distinct contributors. Thanks to everyone who made it happen! #release #opensource #llm #ai

With this new pricing, Terra has just become a real good model to drive WebBrain: www.axios.com/2026/07/30/o...

sending this with kimi k3 on webbrain

We just hit 500 stars on GitHub! Huge thanks to everyone for your interest and support. We hope you're enjoying the project—if you haven't already, please leave a star and help us spread the word! github.com/webbrain-one...

Yesterday was our biggest day ever: 100M tokens served on WebBrain's Cloud alone. 🤗 That's not counting users running their own keys or local models. Wild week.

The ultimate LLM won't be wrapped in an agent framework. The goal is an LLM that is agentic by nature. No wrappers, no external tool-managers—just native agency. #AI #LLM #AGI #AgenticAI #AIagents #AutonomousAgents #MachineLearning #ML #OpenSourceAI

WebBrain just hit a milestone: 1 billion tokens consumed monthly for the first time. The flip side? That would be ~$50K/month in Claude fees. 🥵

We just published a deep dive into the Thinking Machines Inkling benchmark on OpenRouter, exploring the frontier of agentic planning and execution. Check out the full breakdown here: www.webbrain.one/blog/thinkin...

Benchmarked Nanbeige 4.2 3B (BF16) as a browser-agent planner. Surprisingly strong for its size — competitive with much larger open-weight models on WebBrain. www.webbrain.one/blog/nanbeige-42-3b-bf16-planner-benchmark #offlinellm #nanbeige

Benchmarked Poolside's new Laguna S 2.1 (118B-A8B) as a browser-agent planner. 71% Sonnet alignment. Near Hy3/MiniMax M3 but below them, and below Gemma 4 31B QAT / Qwen 3.6 27B (~77%). Best US open-weight in its class, half the size of M3 👇 www.webbrain.one/blog/poolsid... #offlinellm #poolside

Laguna S 2.1 reaches 71% in WebBrain with a high-reasoning request

Poolside Laguna S 2.1 scored 65% with default reasoning and 71% with a high-effort request. It is fast and cheap, but the second run also increased no-tool outputs.

webbrain.one

WebBrain 24.2 (available on github) does support Kimi K3 🎉

WebBrain now supports ChatGPT 5.6 in release 24. It has not hit extension stores yet. Download directly from github.com/webbrain-one...

Major improvements with the new release 23.0.1 Check it out and download yours at github.com/webbrain-one...

Release Release v23.0.1 · webbrain-one/webbrain

Full Changelog: 23.0.0...v23.0.1

github.com

WebBrain is now available on Edge too: microsoftedge.microsoft.com/addons/detai... #WebBrain #MicrosoftEdge #BrowserExtension

Just checked out ThinkingCap and it's really impressive! Great tool for running local AI models. Check out their Qwen36-27B planner benchmark: www.webbrain.one/blog/thinkin...

Is Agents-A1 really better than Qwen 3.6 35B-A3B? www.webbrain.one/blog/agents-... #llm #ai #opensource #qwen #gemma

Agents-A1 enters WebBrain's frozen planner benchmark

Agents-A1 is fast and Sonnet-aligned enough to beat Qwen 3.6 35B-A3B on the headline WebBrain planner metric, but its exact and ideal-name scores are more cautious.

webbrain.one

kids, don't do this at home. we handed our AI browser agent the keys and it: – swiped on Bumble on our behalf – read the NYT past the paywall – auto-replied to strangers at night – ripped a YouTube video – beat a CAPTCHA to sign itself up webbrain.one

WebBrain just hit 150 GitHub stars 🚀 — from a slow burn to a hockey stick in July. Huge thanks to everyone who's starred, contributed, or spread the word. We're building an open-source AI browser assistant. Check it out: github.com/webbrain-one...

🎉 We just hit 100 ⭐ on GitHub! Thank you all for the attention, feedback, and stars — it means the world to us. github.com/webbrain-one...

Meet WebBrain: An Open-Source, Local-First AI Browser Agent That Reads Pages and Automates Tasks in Chrome and Firefox www.marktechpost.com/2026/07/02/m... #localai #localllm #ollama #llm #ai #offlineai #qwen #gemma #llamacpp #lmstudio

Meet WebBrain: An Open-Source, Local-First AI Browser Agent That Reads Pages and Automates Tasks in Chrome and Firefox

WebBrain, an open-source AI browser agent, reads pages and automates tasks in Chrome and Firefox using local LLMs.

marktechpost.com

We take Security seriously at WebBrain #llm #ai #offlineAI #qwen #gemma #gemma4 #llamacpp #openllama #opensource #chrome #ollama #

Hello World!

Ornith 35B claims to beat Gemma4 31B and Qwen3.6 35B on many fronts. We ran it through WebBrain's frozen browser-agent planner benchmark. Result: Ornith is solid — edges Qwen 3.6 on alignment — but Gemma4 remains the better option. webbrain.one/blog/ornith-... #Ornith #Gemma4 #Qwen

Ornith-1.0-35B enters WebBrain's frozen planner benchmark

Ornith's coding-agent claims are impressive. In WebBrain's browser-agent planner benchmark, it lands near the hosted/local planner tier but does not overtake Gemma 4 31B.

webbrain.one