OpenCode Go Frontier Analysis
OpenCode Go Frontier Analysis
Artificial Analysis Intelligence Index v4.3 against the estimated share of OpenCode Go's model-specific monthly usage consumed by one benchmark task. Data pulled 2026-09-26. Sources: Artificial Analysis and OpenCode Go pricing and limits.
The green line is the frontier: after sorting by Go usage share, each point is retained when no cheaper model has a higher Intelligence Index. The x axis is logarithmic because the estimated contributions span several orders of magnitude. Hover points for the model, proxy cost, and quota contribution.
What the percentage means
OpenCode publishes a monthly dollar allowance for each model, rather than a request quota. This chart estimates how much of that allowance one Artificial Analysis task would consume if run through Go's published prices and observed request pattern. A lower percentage means more benchmark-equivalent tasks per month.
The proxy uses OpenCode's documented typical input, cached-read, and output tokens. Cached-write is omitted because the Go usage examples do not specify writes. It is a comparison instrument, not an invoice estimate.
Model detail
| Model | AA model used | II | AA task $ | Go proxy $ | Go allowance $ | Contribution |
|---|
Coverage. The current OpenCode Go documentation lists 33 named models, with peak/off-peak price rows for some models. This page plots the 17 models for which the current Artificial Analysis v4.3 free feed exposes both an Intelligence Index and a cost-per-task value using the exact or closest documented model match. The chart uses the OpenCode Go off-peak price where the documentation gives separate peak and off-peak rows; peak prices are shown in the source table but not combined into a second point. The remaining Go models are retained in the audit but omitted from the graph when AA has no comparable cost-per-task record. DeepSeek V4.1 Flash is now included as a distinct model, matched to AA's DeepSeek V4.1 Flash (Reasoning, Max Effort), II 39.5. It is not the same model as DeepSeek V4 Flash, matched to AA's DeepSeek V4 Flash 0731, II 34.3. Scores, prices, and the Go model list can change; regenerate from the vendored refresh script.