Grok 4.6 pricing for nightly jobs: $0.6667 per 1M on Quanta

01 · What a nightly batch run actually costs on Quanta

A scheduled job is easy to underestimate. One run at 2 a.m. looks trivial on a dashboard, but a team running the same extraction, scoring or summarisation every night multiplies every token by the size of the schedule. Quanta is built for that shape of work, and it makes the cost of each call visible rather than leaving it to a monthly summary line.

Quanta meters every call on real tokens. Input, output, cache writes and cache reads are priced separately, and the dashboard lists the tokens and charge of every individual call. For a nightly pipeline, that means you can trace an expensive run back to the request that produced it instead of reasoning from an aggregate.

Grok 4.6 is listed at $0.6667 for input and $2.01 for output per 1M tokens, with cache reads at $0.1667 and no cache write price listed. Grok 4.7 sits at $0.625 and $1.88, while Kimi K3 is listed at $1.31 and $6.54. For lighter classification or routing steps, Qwen3.7 Plus is priced at $0.1667 for input and $0.4861 for output per 1M tokens. Quanta publishes these figures per model so the nightly budget can be modelled before the job is scheduled.

As worked examples, GPT-6.1 Sol at 10 million tokens a month comes to approximately $4.10, and 100 million to approximately $40.97. Gemini 2.5 Flash works out at approximately $6.49 and $64.93 for the same two volumes. Claude Opus 5.5 is approximately $34.58 and $345.83, and Claude Sonnet 5.5 is approximately $25.07 and $250.69. These are examples at a 3:1 input to output ratio with no cache hits, not quotes for your workload. All listed prices are approximate USD figures, and charges are made in CNY.

02 · Rate limits and concurrency that suit a sustained schedule

Nightly work is sustained work. Quanta publishes and enforces the limits that determine how much parallelism a job can safely use, so you can design the schedule around numbers that will not move underneath you.

• › 120 requests per minute per key.
• › 8 requests in flight per account.
• › 25 MB per request body.
• › 20 live keys per account.
• › Streams have no total-time cap and only end after 300 seconds of upstream silence.

Those limits shape a nightly design directly. A wide fan-out across many documents should be sized against 8 requests in flight per account and 120 requests per minute per key, not against an assumption of unlimited throughput. The 25 MB request body limit is worth checking before a job packs large payloads into a single call. And because streams carry no total-time cap, a long-running generation is not cut off simply for taking time; it ends when the upstream goes quiet for 300 seconds.

Twenty live keys per account gives a data team room to separate workloads, such as one key for the nightly batch and another for interactive checks, without splitting the balance. One key and one balance work across all three protocols, so adding a model family does not mean rebuilding the billing setup.

03 · Matching models to the right client and protocol

Protocol confusion is a common source of wasted overnight runs. Quanta keeps the pairing rules 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.
• › Codex, Cursor, Cherry Studio, Cline, OpenCode, OpenAI SDKs and cURL pair only with GPT, DeepSeek and the other non-Claude, non-Gemini models, including Grok 4.6 and Grok 4.7, Kimi K3, Qwen3.7 Plus, Qwen3.8 Max, Hunyuan 4, GLM-5.3, GPT-5.6 Sol, GPT-6 Astra and DeepSeek V4 Flash and V4 Pro.

For a team running nightly jobs through cURL, an SDK or one of the listed editors, the choice set is the GPT, DeepSeek, Grok, Kimi, Qwen, Hunyuan and GLM families. Quanta exposes them through the OpenAI-compatible endpoint, so an existing integration can point at the Quanta base address with one key and draw on the same prepaid balance.

Claude routes separately, through the Anthropic native protocol from the official Claude clients, and Gemini through its native protocol in Gemini CLI. Getting this right before the schedule starts removes the most common cause of a failed first run.

04 · Prepaid balance, company invoices and per-call visibility

Quanta is pay-as-you-go. There is no monthly fee and no minimum spend; you fund a prepaid balance and draw it down as calls are made. Top-ups are available at $1.39, $6.94, $13.89, $27.78, $69.44 and $138.89, so a team can size an initial balance to a few nights of usage and extend it as the job becomes part of the routine. Sign-up includes $0.50 in trial credit, which is enough to confirm that the endpoint, the key and the client pairing all work before the first scheduled run.

For procurement, company payment is made by bank transfer from a company account, and invoices and advance quotes are arranged through the Contact page or the Messages page of the dashboard. Quanta also offers a straightforward exit: unused balance can be refunded within 15 days of the top-up.

The referral programme pays 19% of a referred account's top-ups and 5% one level further. Rewards can be moved into the balance or withdrawn from $27.78, which is useful for teams that would rather keep the credit inside the workload that generated it.

05 · Data handling that a nightly job can be audited against

Text sent to Quanta is kept for 30 days for abuse handling and billing disputes, then deleted. Model answers are not stored. That retention window gives a team a defined answer for its own governance questions: submitted prompts are retained for a limited, stated period, and generated outputs are not retained by Quanta.

Put the pieces together and the case for Quanta is straightforward for data teams. Published per-token pricing on models including Grok 4.6 at $0.6667 per 1M input tokens, published limits of 120 requests per minute per key and 8 requests in flight per account, a prepaid balance with no monthly fee, and a dashboard that shows the tokens and charge of every call. Protocol rules are stated plainly, so the integration can be configured once and reused every night.

Open an account, add the $0.50 trial credit, point one key at the Quanta base address and run a single representative batch. When the first nightly job completes and the per-call breakdown looks right, the next one is already set up to run without surprises.