OpenCode pricing: how Quanta splits input, output and cache

01 · Why a Quanta usage record shows four separate prices

Most teams open a usage record expecting one number: what did that request cost? Quanta answers a more precise question. Every call is metered on real tokens, and the tokens are sorted into four categories before they are charged: input, output, cache write and cache read. The dashboard lists the tokens and the charge for every call, so the arithmetic behind a bill is visible rather than inferred.

This matters when a team is running sustained, high-volume workloads. A batch job that sends large prompts, a service that returns long completions and an agent that reuses a long context will each behave differently against a single blended rate. Quanta publishes the four rates for every model it lists, and the usage record keeps them apart. That is the difference between a number you can audit and a number you have to trust.

Quanta is pay-as-you-go. There is no monthly fee and no minimum spend, so the cost of a record is the cost of the tokens your team actually sent and received. Sign-up includes $0.50 in trial credit, and top-ups start at $1.39, with further amounts of $6.94, $13.89, $27.78, $69.44 and $138.89 available. Unused balance can be refunded within 15 days of the top-up.

02 · What each of the four token categories represents

Input tokens are the tokens your team sends to the model. Output tokens are the tokens the model returns. These two lines are charged at different rates because they represent different work, and both are recorded on every call.

Cache write tokens and cache read tokens are metered separately again. A record can show tokens written into the cache and tokens served from the cache on the same call, and each is priced on its own line. Because a cache read rate is much lower than a cache write rate in the published table, a workload that reuses context repeatedly can show a visibly different cost profile from the same model handling fresh input every time.

The published rates are per 1M tokens, quoted in $ as approximate USD figures, with charges made in CNY. Take Claude Opus 5.5 as a concrete illustration: input $1.79, output $8.46, cache write $2.11 and cache read $0.0819 per 1M tokens. Gemini 2.5 Flash is listed at $0.2361 input, $1.89 output, $0.2361 cache write and $0.0236 cache read. GPT-6.1 Sol is listed at $0.2083 input, $1.01 output, no cache write rate, and $0.0403 cache read.

Some models in the Quanta catalogue have no cache write rate at all, including GPT-6.1 Sol, GPT-6 Astra, GPT-5.6 Sol, DeepSeek V4 Flash, DeepSeek V4 Pro, Qwen3.7 Plus, Qwen3.8 Max, Kimi K3, Hunyuan 4, Grok 4.6, Grok 4.7 and GLM-5.3. A dash in that position is a fact about the rate card, not a missing record. When a record shows no cache write line, there was nothing to charge at that rate.

03 · How to read one Quanta usage record line by line

Start with the model identifier, then read the four token counts and their matching charges. The order is consistent, so a reader who checks one record can check the next one without relearning the layout.

• › Model: the exact model name and its identifier, such as gpt-6.1-sol or claude-opus-5-5.
• › Input: the number of input tokens and the charge for them.
• › Output: the number of output tokens and the charge for them.
• › Cache write: the number of tokens written to the cache and the charge for them.
• › Cache read: the number of tokens read from the cache and the charge for them.

Because the dashboard shows tokens and charge for every call, a team can sample individual requests rather than relying only on a monthly total. That is useful during a heavy, sustained workload, when a sudden change in cost usually has a cause: a longer prompt, a larger completion, or a shift in how much context is being reused.

Quanta also keeps the record trail in a defined state. The text users send is kept for 30 days for abuse handling and billing disputes, then deleted; model answers are not stored. For a team reviewing spend, that means the record shows what was billed and why, without turning the relay into a long-term archive of your prompts.

04 · What the same record looks like across model families

Quanta speaks three protocols: OpenAI-compatible at /v1, Anthropic native and Gemini native. One key and one balance work across all three, and the usage record is formatted the same way regardless of which route a team uses.

Client pairing is explicit. Claude models work only from the official Claude Code and Claude Desktop clients, with Claude Code 2.1.285 or newer. Gemini models use the native protocol only, through Gemini CLI. OpenCode pairs with the GPT, DeepSeek and other non-Claude, non-Gemini models in the catalogue, including Qwen3.7 Plus, Qwen3.8 Max, Kimi K3, Hunyuan 4, Grok 4.6, Grok 4.7 and GLM-5.3.

That pairing shows up directly in the pricing a record carries. An OpenCode team working on GPT-6.1 Sol will read input $0.2083 and output $1.01 per 1M tokens, with cache reads at $0.0403. A team working on Claude Opus 5.5 through Claude Code or Claude Desktop will read input $1.79, output $8.46, cache write $2.11 and cache read $0.0819. A Gemini 2.5 Flash team on Gemini CLI will read input $0.2361, output $1.89, cache write $0.2361 and cache read $0.0236. The record format does not change; the rates do.

Published examples, using a 3:1 input to output ratio with no cache hits, help put the four lines in context. For GPT-6.1 Sol, 10 million tokens a month is approximately $4.10 and 100 million is approximately $40.97. For Gemini 2.5 Flash, the same volumes are approximately $6.49 and $64.93. For Claude Opus 5.5, they are approximately $34.58 and $345.83. These are examples, not quotes or commitments, and they assume no cache activity.

05 · Why teams can budget on Quanta without reading between the lines

Cost clarity only helps if the surrounding terms are also clear. Quanta publishes and enforces its limits: 120 requests per minute per key, 8 requests in flight per account, 25 MB per request body and 20 live keys per account. Streams have no total-time cap and only end after 300 seconds of upstream silence. A team running a heavy pipeline can therefore reason about throughput from stated numbers instead of trial and error.

For companies, Quanta supports bank transfer from a company account, and invoices and advance quotes are arranged through the Contact page or the Messages page of the dashboard. There is no monthly fee, no minimum spend and no plan to buy into, so spend follows usage.

Referral rewards add a further layer: 19% of a referred account's top-ups, and 5% one level further, with rewards withdrawable from $27.78 or movable into the balance. And because every call is broken into input, output, cache write and cache read, the same statement applies to the last dollar as to the first: it is metered on real tokens, and it is visible on the record.

Open an account at Quanta, add a balance from $1.39, and read your first usage record with all four lines in front of you. For teams with heavy, sustained workloads, that is where good value becomes checkable.