← Voltar aos guias

GPT-6.1 Sol pricing: input vs output tokens on 59API

If you have ever opened a usage dashboard and felt that the total did not match the mental math you did before the request, you are not alone. The gap usually comes from one thing: input tokens and output tokens are priced very differently, and almost every prompt you send is mostly input. 59API is an independent third-party AI API relay — one account, one API key, one API host at https://api.59api.com — for Claude, GPT, Gemini, DeepSeek and other models. Billing is prepaid, charged per request by real token usage, with no monthly fee and a balance that never expires. This checklist walks through the numbers so you can see what a request actually costs before you make it, starting with gpt-6.1-sol and working outward.

Why the input and output split decides your bill

Every request has two sides. Input tokens are what you send: system prompt, conversation history, retrieved context, tool schemas. Output tokens are what the model generates: the reply, the code, the JSON. On 59API the two sides are billed at different rates per 1 million tokens, and output is consistently the expensive side. Take gpt-6.1-sol: input is $0.1847 (¥1.33) per 1M tokens while output is $0.9264 (¥6.67) per 1M tokens. gpt-5.6-sol shows the same shape at a different scale — input $0.3472 (¥2.5), output $1.94 (¥14). So a prompt-heavy workflow is cheaper than a generation-heavy one, even at identical token counts. There is also a third line item worth knowing: cache read. Re-reading cached context costs $0.0375 (¥0.27) per 1M tokens on gpt-6.1-sol and $0.0278 (¥0.2) on gpt-5.6-sol, which is well below fresh input. 59API bills all three lines from real token usage, so the shape of your workload shows up directly in the charge.

Worked examples: real numbers, not round guesses

Here is where the split stops being abstract. Consider gpt-6.1-sol at input $0.1847 (¥1.33) and output $0.9264 (¥6.67) per 1M tokens. A hypothetical run of 5M input plus 1M output tokens without cache comes to $1.85 (¥13.32) on 59API. At the official list price of $2 / $10 per 1M tokens that workload would be $20 (about ¥144), so 59API lands at about 9% of list price. Now take gpt-5.6-sol with 2M input and 0.5M output tokens without cache: that is $1.67 (¥12) on 59API, against $25 (about ¥180) at official list price — about 7% of list price. On the Claude side, claude-sonnet-5-5 at input $1.11 (¥8) and output $5.56 (¥40) costs $6.11 (¥44) for 3M input plus 0.5M output tokens without cache, versus $16.5 (about ¥118.8) at official list price, or about 37% of list price. Note that official list prices exist only for Claude and GPT models; they are a reference price, nothing more.

Cache read: the line item that rewards repetition

Agentic loops resend the same context over and over — the system prompt, the file tree, the tool definitions. That is where cache read earns its place in your budget. On gpt-6.1-sol, cache read is $0.0375 (¥0.27) per 1M tokens against $0.1847 (¥1.33) for fresh input. On claude-sonnet-5-5 it is $0.0972 (¥0.7) versus $1.11 (¥8) input, and on claude-opus-5-5 it is $0.0694 (¥0.5) versus $1.53 (¥11). The pattern holds across the catalog: deepseek-v4-flash reads cache at $0.006944 (¥0.05) per 1M tokens versus $0.2778 (¥2) input, and MiMo-V2.6-Pro reads at $0.002778 (¥0.02) versus $0.2083 (¥1.5). If your workload reuses context, the cache line can become the majority of your input spend — at a fraction of the fresh-input rate. 59API reports cache reads as their own usage line, so you can see exactly how much of your input traffic was re-read rather than newly ingested. Worth checking in any request where the same prompt prefix appears repeatedly.

What one request looks like on 59API

Connecting is a two-line change, and which client you use decides which models you can reach. For gpt-6.1-sol and the rest of the GPT family, along with DeepSeek and other non-Claude, non-Gemini models, use an OpenAI-compatible client such as Cursor, Cherry Studio, Cline or OpenCode, or Codex and the OpenAI SDKs: base_url https://api.59api.com/v1, chat endpoint https://api.59api.com/v1/chat/completions. Claude models work only through the genuine Claude Code or Claude Desktop client (2.1.285 or newer) with ANTHROPIC_BASE_URL=https://api.59api.com and ANTHROPIC_AUTH_TOKEN=YOUR_59API_KEY — never through the OpenAI SDK, cURL or other clients. Gemini models use Google's native protocol only, via Gemini CLI with GOOGLE_GEMINI_BASE_URL=https://api.59api.com and GEMINI_API_KEY=YOUR_59API_KEY; they are not OpenAI-compatible. Once connected, each request is billed from your prepaid balance by real token usage. Top-ups run from ¥10 to ¥1000 via WeChat Pay or bank card, invoices are available, and billing is in CNY with USD figures shown at ¥7.2 = $1.

Optional cards, if you want a fixed runway

If you prefer a capped budget over a wallet, 59API offers optional cards. A card has its own key, calls only that family's models, and is paid only from the card — it never falls back to wallet balance, and a balance key never draws on a card. Unused quota does not roll over, and quota and caps are expressed in official-price USD. A concrete example: gpt-6.1-sol and the rest of the GPT family on the G-PRO Starter card (30 days) cost $83.19 (¥599), with quota $3250 and a weekly cap $1250. Quota drains at official price × 2.4, so that card covers about $1354.17 of usage at official list prices. The Claude-side C-PRO Starter card (30 days) is $83.19 (¥599) with quota $1300, weekly cap $500, and about $541.67 of usage at official list prices on the same × 2.4 drain. Shorter options exist too, like the 7-day G-PRO $1500 card at $40.56 (¥292). Whether you pay per request or run on a card, the input-versus-output math above still tells you what your workload is worth.

Docs