OpenCode Go's New Pricing, Explained

OpenCode Go changed how its $10/month plan works. The old system counted requests per model in a rolling 5-hour window, with higher tiers to raise the ceiling. The new system, live on opencode.ai/go, meters usage in dollars at each model's published per-token rate — and there's now exactly one plan.

If you're picking a coding plan, this update makes Go much easier to reason about, and it quietly changes who it's for. Here's the OpenCode Go pricing model in full, what a dollar actually buys, and where the ceiling sits.

What Changed

Three things:

1. Request counts became dollar metering. Every request is billed against your allowance at the model's listed input/output rates — the same per-million-token math an API bill uses, just prepaid.
2. Tiers became one plan. $10/month, no higher tiers, no annual option, one subscriber per workspace.
3. The catalog crossed the open-weight line. Alongside the open coders — GLM-5.2, Kimi K3, Qwen3.8 Max, DeepSeek V4, MiniMax M3, MiMo V2.5 — Go now carries its first closed models: Grok 4.5 and GPT-5.6 Luna.

The Limits, Exactly

The $10 plan stacks three time-window caps:

On top of that, each model carries its own monthly allocation:

Because metering runs at per-token rates, how far $60 goes depends entirely on which model you pick. At Grok 4.5's rates ($2/M input, $6/M output) the $15 allocation is a handful of serious agent sessions. At MiMo V2.5's rates ($0.14/M in, $0.28/M out) the monthly ceiling is, for most people, effectively unreachable.

Past a limit, requests drop to the free models — currently Ox Alpha Free, explicitly marked limited-time — or, if you enable the Use balance option, Go draws on your Zen pay-as-you-go balance instead of blocking.

The Math per Dollar

Here's the part worth being honest about: $10 for up to $60 of usage is a 6× ratio, and that's the best ratio of any flat coding plan we know of — including ours. DevPass turns every $1 into roughly $3 of usage at provider rates.

The difference is what the ratio is attached to. Go's 6× ends at $60 of included usage — you can keep going from an opted-in Zen balance, but that's your own money at 1×. There is no bigger plan to move to, and the per-model allocations mean the models you'd most want for hard problems — Grok 4.5, GPT-5.6 Luna, Kimi K3 — stop at $15 each. DevPass's 3× is attached to allowances of ~$87 (Lite, $29), ~$237 (Pro, $79) and ~$537 (Max, $179), with Claude Opus 5, GPT-5.6 Sol and Gemini 3.1 Pro in the catalog and opt-in pay-as-you-go overflow past the cap.

So the decision reduces to two questions:

• Does your month fit inside $60 of usage? Count agent sessions honestly — a single long Claude Code-style run can burn several dollars on a mid-priced model.
• Do you ever need Claude, Gemini, or frontier-tier GPT? Go has none of them. GPT-5.6 Luna is OpenAI's cost-optimized variant, not the frontier Sol tier.

If your month fits inside $60 of usage and you don't need Claude, Gemini or frontier-tier GPT, take Go — it's the cheapest seat in the market. If either condition fails, choose the bigger allowance: the ceiling, not the price, is what you're actually choosing between.

Where Go Fits Now

The update repositions Go from "cheap requests for open models" to something closer to a starter meter: transparent dollar pricing, a taste of the closed frontier, and a hard monthly stop that keeps it at $10. That's a genuinely good product for evenings-and-weekends coding on open-weight models.

The full head-to-head — catalog, caps, overflow, dashboards — is in our updated DevPass vs OpenCode Go comparison. For the wider field, see the best AI coding plans in 2026, where Go now appears in the ranking.

• Try DevPass — flat plans from $29/mo with ~3× usage across 200+ models, Claude and Gemini included
• Compare DevPass vs OpenCode Go — the detailed head-to-head
• Read the full coding-plan ranking — eleven plans, compared honestly