OpenCode setup on Vana API with $0.10 in sign-up credit

§ One address, three protocols, minimal setup

The first question OpenCode users tend to ask is how much of their setup has to change. With Vana API, the answer is short. Model requests go to a single address, https://api.vanaapi.com, and that address answers three protocol styles. The endpoint stays the same; the protocol and the model id are what you select for the work you are doing.

For OpenCode, the relevant protocol is the OpenAI format at /v1/chat/completions. The models listed in the OpenAI-format group can be reached from OpenCode, and the same format is used by Cursor, Cline, Roo Code, Cherry Studio, any OpenAI SDK or cURL. GPT models additionally answer on /v1/responses, which is what Codex uses. The other two protocol styles on the Vana API address are reserved for the clients that speak those protocols directly, so each model is reached through the client that matches its format rather than through a universal adapter.

It is worth keeping two addresses clearly in mind, because this is where most confusion starts. The address https://api.vanaapi.com only answers API requests. The dashboard, where keys are created, balance is shown and the usage log is kept, is on the main site, https://vanaapi.com. The working pattern is therefore small: create a key in the dashboard, point OpenCode at the API address, and choose a model id from the OpenAI-format list.

There is no monthly commitment in that flow. Vana API uses a prepaid balance, charges per real token and has no monthly fee. New accounts receive $0.10 of credit on sign-up, which is enough to make a small test request and confirm that the key, address and model id are correct before adding a larger amount.

§ Which models can OpenCode reach through Vana API?

OpenCode pairs with the OpenAI-format group on Vana API: GPT, Grok, DeepSeek, GLM, Kimi, Qwen, MiniMax and MiMo models. The practical benefit is that you can compare output or cost across that range while keeping one client and one endpoint. The model id is what selects the route, so it is worth copying it exactly rather than guessing at a shortened name.

Some examples you can select in OpenCode today. GPT-5.6 Sol uses the model id gpt-5.6-sol. GPT-6 Astra uses gpt-6-astra. GPT-6.1 Sol uses gpt-6.1-sol. On the DeepSeek side, DeepSeek V4 Flash is deepseek-v4-flash and DeepSeek V4 Pro is deepseek-v4-pro. Other options include GLM-5.3 at glm-5.3, Kimi K3 at kimi-k3, Qwen3.7 Plus at qwen3.7-plus, Qwen3.8 Max at qwen3.8-max, MiniMax M3 at MiniMax-M3 and MiMo V2.6 Pro at MiMo-V2.6-Pro. Grok 4.7 uses grok-4.7.

Moving between these models in OpenCode does not mean rebuilding the connection. The base address, the key and the OpenAI-format path stay as they are, and only the model id changes. That is the practical value of the one-address layout: you are not maintaining a separate endpoint for each family of models, and you are not writing a custom adapter for each one.

For example, say you run a short daily coding task in OpenCode and want to compare a GPT model with a DeepSeek model on the same prompt. You would change the model id in the client, keep the same Vana API key, and read the results in the usage log on the dashboard. The example is deliberately ordinary, because that kind of comparison is what the setup is meant to support.

§ How do the prepaid balance and usage log work?

Balance questions from OpenCode users usually come down to visibility. With Vana API, you top up a prepaid balance and spend from it as calls are made. There is no monthly fee and no subscription to cancel. Each call is charged according to the real tokens it consumes, so the balance reflects actual usage rather than an estimate.

The dashboard is where you confirm this. Keys are created there, the current balance is displayed there, and every call is listed in the usage log there. When you are testing a model id or comparing prices, you can check the log after a request and see what actually happened. For an OpenCode user, that removes a common blind spot: the client shows you the conversation, while the dashboard shows you the billing side.

Because billing is pay-as-you-go, a sensible habit is to start small. The $0.10 of credit on sign-up gives you room for a quick check. If you then want a larger working balance, you can top up and watch the usage log line by line. There is no need to guess which commitment will fit, because a monthly plan is not required in the first place.

One detail to keep in mind is that the API address and the dashboard address do different jobs. A dashboard request to https://api.vanaapi.com will not return your balance, and model requests are not served from the main site. Keeping the two straight accounts for most of the setup work.

§ Can unused balance be refunded?

Yes, within a defined window. Unused balance can be refunded within 3 days of the top-up, minus what was used and the payment processor's fee. That is the full policy, and it is worth reading before adding a large amount during a busy week of OpenCode work.

The refund is calculated on what is left, not on what you originally paid. If you top up an amount and then make calls that consume part of it, the spent portion is deducted and the payment processor's fee is deducted as well. The remainder is what can be returned. The closer a top-up sits to the point where you still need it, the more useful that window is.

A practical routine for OpenCode users is to top up for the work you expect rather than for a distant scenario. Say you plan to test several models over a few days of coding tasks; topping up for that period keeps more of the balance inside the refund window, while topping up far ahead of time moves more of it outside it. This is an example of how the policy behaves, not a statement about any particular amount.

Refunds and referral earnings are separate questions, and it helps to keep them apart. The refund applies to unused balance within 3 days of a top-up. Referral earnings are withdrawn from a different threshold, described below. Neither affects the other.

§ What do referrals pay, and when can earnings be withdrawn?

Vana API runs a two-level referral programme. You receive 14% of what people you invite pay, and 5% one level further. The percentages apply to what the invited users actually pay, so the programme follows real payments rather than sign-ups alone.

Withdrawal starts at $13.89. Once referral earnings reach that point, they can be withdrawn. For OpenCode users who already spend on API access, sharing the endpoint with a colleague or a small group is the direct route to that threshold, because the percentage is tied to what those people pay.

It helps to be precise about what the referral programme does. It pays a share of what invited people pay, at 14% for your direct level and 5% one level further. It does not change your own token pricing, and it sits separately from the refund window on unused top-up balance. If you are unsure how a given payment should count, the dashboard remains the place to check, since that is where balance and usage records are kept.

Taking the pieces together gives a clear picture: prepaid balance, pay-as-you-go billing, per-call usage records, a refund window of 3 days, and referral earnings that can be withdrawn from $13.89. None of that requires extra infrastructure in your OpenCode setup.

§ What do the token prices look like for OpenCode?

Prices on Vana API are listed per million tokens, as input and output. For the GPT models, the comparison against official list pricing is straightforward. GPT-5.6 Sol is $0.3472 / $1.81 on Vana API against an official list price of $4 / $20. GPT-6 Astra is $0.6944 / $3.54 against $10 / $50. GPT-6.1 Sol is $0.1389 / $0.8333 against $2 / $10.

The other OpenAI-format models carry the same per-million-token format. Grok 4.7 is $0.4861 / $1.53. DeepSeek V4 Flash is $0.2778 / $1.04, and DeepSeek V4 Pro is $1.11 / $3.33. GLM-5.3 is $0.5556 / $1.67. Kimi K3 is $1.04 / $5.21. Qwen3.7 Plus is $0.1278 / $0.4167, Qwen3.8 Max is $0.6944 / $2.08, MiniMax M3 is $0.1389 / $0.5556, and MiMo V2.6 Pro is $0.2083 / $0.3472.

For teams that prefer a fixed monthly arrangement, there is one plan: GPT up to $2,750 / 30 days for $103.61 for 30 days, at most $820 a week, covering GPT models only. That plan sits alongside the prepaid pay-as-you-go option, so you can pick the arrangement that fits how you work.

Put together, the picture for an OpenCode user is deliberately plain. One address, three protocols, minimal setup. Keys from the dashboard, prepaid balance, per-token charging, a usage log for each call, refunds on unused balance within 3 days of the top-up, referral earnings withdrawable from $13.89, and published per-million-token pricing for the OpenAI-format models OpenCode can reach. That is the surface you work with.