#Free?
hard marten
openrouter has free models.
build.nvidia.com has free LLM APIs
if you sign-up for many of the agentic cli's like gemini, codex, opencodee, kilo they have free API tokens
Many of the agentic IDEs have free API tokens: Antigravity, VSCode Insiders, Kilo, etc.
opencode go is a good subscription service with $5/month subscription for the first month that increases to $10.
Alibaba code has a openclaw/hermes style subscription for $50/month
My experience is if you want to get value go with $20/month Minimax
flat sky
Open router qwen 3.6 plus is free?
hard marten
You need three models to run Honcho:
- Embedder - Gemini-1.5-flash-lite will be the cheapest and most effective. you can set a budget of just a few dollars for embedding.
- Deriver - You want to use a good reasoning model to derive associations, so an Nvidia API that uses a GLM5 or Kimi-K2.5
- Reasoner - NVIDIA again or if you get some Opus/Gemini High 3.1, etc.
There's always tons of free models that are in trial period. Just float between them constantly jumping from openrouter free to Nvidia free to kilo to opencode to antigravity
celest sphinx
My Hermes said
For honcho local
Sounds good to me. Excellent backup combo:
Main (free):
• qwen/qwen3.6-plus-preview:free
• minimax/minimax-m2.5:free
• z-ai/glm-4.5-air:free
Embeddings:
• qwen/qwen3-embedding-8b
Backup:
• nvidia/nemotron-3-super-120b-a12b:free
• stepfun/step-3.5-flash:free
6 models in total, all free. Nemotron has 262K context and Step 256K, both with tool calling. A good combination of different providers (Qwen, MiniMax, Z-AI, Nvidia, StepFun) so if one goes down, the others keep working.
Final configuration summary:
Dialectic reasoning: qwen3.6-plus-preview:free (all levels)
Deriver: minimax-m2.5:free
Summarization: glm-4.5-air:free
Embeddings: qwen3-embedding-8b ($0.01/M)
Backup: nemotron-3-super + step-3.5-flash
Shall we proceed with the Docker + Honcho installation? I need your permission to install Docker.