Use case
Coding agent
A model that can hold a repo in context, call tools, and edit code.
What matters
Tool support and context length decide whether an agent can work. Leaderboard rank changes weekly — check Arena's coding tab and Artificial Analysis before you lock a stack.
accuracy · long-context · structured-output
Top picks
- #1ricursiveAI/Ripple-65B-Preview
Listed at a low input price (under $1 / M), which is what bulk work cares about.
- #2NVIDIA: Nemotron 3.5 Super VL (free)
Open-weight hint on the listing, so you can check the license and run it yourself.
- #3Tencent: Hy4
Open-weight hint on the listing, so you can check the license and run it yourself.
Budget pick
NVIDIA: Nemotron 3.5 Super VL (free)
Open-weight pick
NVIDIA: Nemotron 3.5 Super VL (free)
Pitfalls
- A decision model will not write the patch. It can grade one, or route the ticket to the agent.
- The cheapest chat model with tools is not automatically the one that should merge code.
Sample prompt
You are a coding agent. Read the failing test, name the file you will change, then make the smallest edit that passes.
Try the chat picks