OpenCode Go Limits by Plan: Dollar Allowances per Model on Go, Go Plus and Teams

Monthly limit per model on each plan

The dollar figures are OpenCode's monthly limits, and token prices are the same on all three plans. The last column is OpenCode's own estimate for Go, which assumes a typical coding-agent request that rereads a large cached context.

Read the last column, not the dollar columns: the same $15 buys about 4,230 GPT 6 Luna requests per 5 hours but 147 on Kimi K3.1 Pro, because Kimi's tokens cost about 20 times as much and OpenCode assumes larger Kimi requests. Go's 30 models also include Kimi K3, GLM-5.2, GPT 5.6 Luna, Qwen3.8 Max and Flash, MiniMax M3.1 Flash, LongCat-2.0, Step 5 Preview, Hy4 preview, Hy3, three more MiMo models and two Muse Spark Contributor models, plus two models that are free for a limited time. OpenCode says the list may change as it tests new ones. Gemini 4 Argon's token price is half off through November 30, 2026, and OpenCode raised the DeepSeek V4.1 Flash limits on October 30, 2026; OpenCode says limits may change.

The 20% and 50% caps decide your day, not the monthly total

Because the caps are shares of each model's monthly limit, the 5-hour cap on Go is $20 of DeepSeek V4.1 Flash, $12 of the $60 models and $3 of the $15 models. A week's cap is half the month, so you cannot spend a whole month's limit in one week. OpenCode estimates Go's Kimi K3.1 Pro allowance at about 147 requests per 5 hours, 368 per week and 735 per month: filling the 5-hour window two and a half times uses up the week.

OpenCode says each model's limit determines how its usage counts toward these allowances, and every response reports how much it counted as opencode.allowance. It does not spell out whether time on one model reduces what is left on another, so check that field before planning to stack several models' limits.

Go Plus quadruples the $15 models but only doubles DeepSeek V4.1 Flash

Go Plus costs four times Go. It raises the limit four times on most $15 models and eight times on GLM-5.3. On the $60 models it is three times for five of them (GLM-5.3-Flash, GLM-5.2, Kimi K2.7 Code, MiniMax M3 and Qwen3.7 Plus), four times for LongCat-2.0 and Hy3, and only twice for MiMo-V2.6-Flash, MiMo-V2.5 and the two Muse Spark Contributor models, as it is for DeepSeek V4.1 Flash. Even the weakest case beats paying per token: in ToolColumn arithmetic, Go Plus's extra $30 adds $100 of DeepSeek V4.1 Flash usage, $45 on a $15 model and $60 to $180 on a $60 model, while $30 of pay-as-you-go top-ups buys about $28.40 of usage after the processing fee. If you hit one model's limit every month, upgrade before turning on Extra Usage.

At the limit: free models, or Extra Usage on your balance

When a model's limit is used, Go stops serving it until the window rolls forward. You can switch to OpenCode's free models, which are offered for a limited time and change. If you have pay-as-you-go credit and turn on Extra Usage, Go instead falls back to that balance at OpenCode's pay-as-you-go prices, which for DeepSeek differ from Go's: V4 Pro costs $1.74 input and $3.48 output per 1M tokens, and V4.1 Flash has no off-peak rate. The balance is bought in prepaid top-ups with a 4.6% plus $0.31 processing fee, so $50 of credit costs $52.61. Go Teams starts with Extra Usage on.

DeepSeek models need an admin to allow the Global region

Go's workspace privacy settings allow only Europe and United States regions by default, and block paid endpoints that train on request data. DeepSeek V4.1 Flash and V4 Pro are blocked until an owner or admin selects Global under Regions in Settings, Privacy; workspaces that never saved privacy settings already allow Global for requests from China. The two Muse Spark Contributor models need "Allow paid endpoints that train on request data" turned on. OpenCode lists a zero-data-retention agreement with DeepSeek that it renews monthly.

Go Teams: each member's own limits, $20 at any point in the month

Go and Go Plus are for one-member workspaces. Go Teams covers every member of a shared workspace at $20 each, and each member gets their own limits at Go's level plus $5 a model. A member who joins mid-period adds $20 to the workspace's bill and gets a full month's allowance, and service accounts are not covered. Moving from Go or Go Plus to Teams credits the unused part of the month.

Go also works in Claude Code, Codex and other agents

Go is an API as well as an OpenCode provider. Its endpoint at opencode.ai/inference/go speaks the OpenAI, Anthropic and Gemini formats, and OpenCode lists Claude Code, Codex, Hermes, Pi, ZCode, jcode and Kilo Code CLI as validated clients and documents a Raycast extension. Clients must send a stable session ID so Go can route and cache requests, and OpenCode monitors traffic for abuse. It lists DeepSeek Harness, GitHub Copilot Chat, Kimi Code and MiMo Code as problematic because their session headers are missing or incomplete. The limits are the same whichever client sends the request.