OpenCode vs Claude Code: models, pricing and cost, measured on one machine

OpenCode is an open-source coding agent that runs models from more than 75 providers; Claude Code is Anthropic's and runs Claude. The model you choose decides most of the cost. On my machine, which has run both, the median OpenCode call cost $0.002, mostly on DeepSeek V4 Pro, and the median Claude Code call $0.094 — though the OpenCode calls carried more context.

OpenCode pricing vs Claude Code pricing

OpenCode costs nothing itself. You pay for the model, three ways: your own API key for a provider — DeepSeek, OpenAI, Anthropic and so on — billed by that provider; OpenCode Zen, a pay-as-you-go gateway to tested models, billed per request from prepaid credit, with a few models free; or OpenCode Go, a subscription to mostly open models plus Claude Haiku 5.5, whose dollar caps per model reset over five hours, a week and a month. Every Go plan, the Zen rates and what an OpenCode call costs: OpenCode pricing.

Claude Code comes with a Claude plan or bills an API key per token. Every plan and its limits: Claude Code pricing.

Can I use my Claude Code subscription (Pro or Max) in OpenCode?

Not legitimately. OpenCode's providers page says plugins that let OpenCode use a Claude Pro/Max plan exist, that "Anthropic explicitly prohibits this", and that OpenCode stopped bundling them in version 1.3.0. Claude Code's own documentation reserves subscription sign-in for ordinary use of Claude Code and Anthropic's other native apps, and says third-party developers may not route requests through Free, Pro or Max credentials on behalf of their users.

So Claude inside OpenCode is billed per token — an Anthropic API key, or Zen, which lists Claude at the same rates as Anthropic's list: Opus 5.5 at $4 input and $20 output per million. What a Claude call costs at those rates, model by model: Claude API pricing.

What each cost on one machine

Caprock reads OpenCode's database and Claude Code's transcripts into the same table. OpenCode's figures are its own — it computes the cost and Caprock passes it through; Claude Code's are priced at Anthropic's list prices. A call is one request to the model, counted the same way on both sides. Read on 3 October 2026: 9,309 OpenCode calls on 21 days in May and June 2026, 84% of them on DeepSeek V4 Pro and 16% on MiniMax M3, and 79,014 Claude Code calls on 56 days, 18 July to 2 October 2026.

The OpenCode calls cost less while carrying more context — 298K tokens at the median against 169K — because of the price of the cache. Both agents re-read almost everything from it, and DeepSeek V4 Pro lists a cache hit at $0.022–$0.044 per million where Opus 5 lists $0.50. Per million tokens of every kind, the DeepSeek calls came to $0.05 and Claude Code's Opus 5 calls to $0.63.

What this does not show: which model writes better code, or what the same task costs in each. The two ran in different months, on different projects, with different models, and DeepSeek V4 Pro is not Opus 5. It is what the work I actually gave each one cost, read the same way.

Which one to pick on cost

If you want Claude and use it all day, Claude Code on a Max plan: in OpenCode, Claude bills per token, and a median Claude Code day of mine came to $153 at list prices. If you want to choose the model per task, OpenCode: the bill follows the model, and on a cheap open model my median OpenCode session cost four cents. If you run both, add their usage up in one place to see the total.

Codex vs Claude Code makes the same comparison with OpenAI's agent.

Measure both on your machine

Caprock finds OpenCode through its own database, opened read-only, and Claude Code through its transcripts — nothing to configure, and sessions from before you installed it are included. Caprock's Home screen, its live view, filters by agent, the Cost screen adds both up, and caprock export writes every call of both:

$ caprock export events --format tsv --out events.tsv

Claude Code's dollars are a list-price estimate, as its own /usage is, beside the plan's dollars and the share of the weekly limit the work used; OpenCode's are its own figure. Free, open source; nothing leaves your machine. Install Caprock and it reads back every session both agents still have on disk, started in any terminal.