Or you can use Gemini 3.7 Flash as the worker model https://
github.com/Shubhamsaboo/a
wesome-llm-apps/tree/main/agent_skills/advisor-orchestrator-worker
…
@saboo_shubham_
-

Gemini 3.7 Flash as a Worker in Multi-Model Agents
By
–
-
Open-Source Templates for AI Agents and RAG Apps
By
–
GitHub Repo: https://
github.com/mvakde/mdlARC/ Open-source Awesome LLM Apps repo (135k+ stars) and 100+ AI Agents, Agent Skills and RAG templates: https://
github.com/Shubhamsaboo/a
wesome-llm-apps
… -

A Small Transformer Reports 44% on ARC-AGI-1 After Brief Training
By
–
Crazyy…this guy hit 44% on ARC-AGI-1 in just 67 cents. By training a small transformer from scratch in 1.5 hrs on a 5090. 100% open-source code.
-
Agentic video understanding with Gemini 3.7 Flash is HUGE
By
–
Agentic video understanding with Gemini 3.7 Flash is HUGE.
— Shubham Saboo (@Saboo_Shubham_) 1 septembre 2026
Just reduced cost, less tokens, and higher accuracy. No biggie. pic.twitter.com/zEF9B6c4HTAgentic video understanding with Gemini 3.7 Flash is HUGE. Just reduced cost, less tokens, and higher accuracy. No biggie.
-
Agent Skill Benchmark Builds a Separate Knowledge Wiki for Each Dataset
By
–
Task-specific. The wiki is built per dataset, not shared across benchmarks. From the setup: the skill set starts empty and is "developed for each dataset through the evolution process."
-

Persistent Wikis Help Smaller Agents Beat Larger Models Without Skills
By
–
MUST READ paper from Google Research. Interesting bits… 1. Agents get better at tasks by keeping a wiki of everything they've learned. 2. Small model with skills beats one 3x its size without them. 3. Delete the wiki and the gains go away.
-

Hermes Agent Organization Combines Human Direction, Specialist Roles and Persistent Learning
By
–
How i run a self-improving Hermes agent organization built on four pillars. • Direction: I set goals, approve, and correct when needed. • Execution: Hermes chief-of-staff, specialists agents own domains, and CLIs act.
• Learning: Skills preserve what works. Memory stores -

Uber Reports Agent-Generated Pull Requests and Falling Compute Costs
By
–
Uber published what its agents actually cost. 70%+ of pull requests now come from agents. 3,600 agent skills, 30K skill runs a day. Cost per 1,000 requests for a given frontier model has fallen 34% from its peak, and cost per session is down 52%.
-
Open-Source LLM App Repository Includes Agent and RAG Templates
By
–
GitHub Repo: https://
github.com/experientialla
bs/experiential
… Open-source Awesome LLM Apps repo (135k+ stars) and 100+ AI Agents, Agent Skills and RAG templates: https://
github.com/Shubhamsaboo/a
wesome-llm-apps
… -

Open-Source Agent Gateway Routes Hosted and Local Models Through One API
By
–
This is open source gateway and router for agents. – one api for hosted, BYOK and local models
– controls dashboard – model router optimize for cost, speed and quality And you pay $0 extra for all these controls.