Skip to content

models.md

Latest commit

 

History

History
50 lines (36 loc) · 2.74 KB

models.md

File metadata and controls

50 lines (36 loc) · 2.74 KB

Models — free opencode zen vs. GPT / Claude

opencode itself does not ship a proprietary model. It routes to 75+ LLM providers via Models.dev. The fastest way to try opencode without a credit card is opencode zen — opencode's hosted gateway with a rotating free tier alongside paid models.

This page focuses on the free zen models because the workshop is about trying the tool. We compare them to the dominant paid models from OpenAI and Anthropic so you know when it is worth reaching for a paid key.

Free opencode zen models

Model Best for
Big Pickle General agentic coding — good default first pick
DeepSeek V4 Flash Free Fast iteration, lightweight edits
MiMo V2.5 Free Reasoning-heavy refactors
North Mini Code Free Snappy code-completion-style tasks
Nemotron 3 Ultra Free Long-context analysis when latency is not critical

The free roster rotates. Always confirm what is currently free at https://opencode.ai/docs/zen/ before the workshop.

Free zen vs. GPT / Claude — at a glance

Dimension Free zen models GPT-5.x Claude 4.x (Opus / Sonnet)
Cost $0 (rate-limited, may change) ~$1.25–$10 / 1M in, ~$10–$30 / 1M out ~$3–$15 / 1M in, ~$15–$75 / 1M out
Code quality Good for everyday edits; weaker on subtle refactors Excellent across breadth Excellent, especially on multi-file refactors
Tool-use reliability Usable but expect occasional retries Very reliable Very reliable
Context window Varies per model (typically 32k–128k) 200k+ 200k+
Latency Fast (DeepSeek Flash) to slower (Nemotron Ultra) Medium Medium–slow on Opus, medium on Sonnet
Privacy Hosted by opencode zen Hosted by OpenAI Hosted by Anthropic
Local option Pair with Ollama / LM Studio for fully-local runs No No

How to switch

Two ways to change model from inside opencode:

  • Press f2 (recent) or <leader>m (ctrl+x then m) to open the picker. See commands.md.
  • Set a default in opencode.json:
{
  "model": "opencode-zen/big-pickle"
}

Format is always provider_id/model_id. To pin a paid model you would write, for example, "anthropic/claude-sonnet-4-5" or "openai/gpt-5".

Workshop guidance

  • Start free. Use big-pickle for the challenge to feel the loop without burning a paid quota.
  • Compare deliberately. During challenge development, run the same prompt on different free models — read both diffs, discuss.
  • Watch the rate limits. Free tiers throttle quickly with multiple students hitting them. Have one paid key ready as a backup.