Checking back in on the state of open weight LLMs after a couple of months away and the best are much better than in Jan – notably gemma4 and qwen3.5.
Unlike AI agents like OpenClaw, it’s pretty safe to play with them at home using tools such as Ollama and AnythingLLM.
Warning: using them may make you desperately yearn for more CPU and RAM in a way I only dimly remember from the late 90s.
Governments and large orgs should be offering internal product teams free and easy access to hosted versions of such open models since they radically reduce the cost and vendor dependency vs cloud Claude, OpenAI, CoPilot etc. And they’re plenty good enough already for much stuff.
Play. Play with this stuff.
Tom Loosemore 
Leave a comment