Gemini 3.8 Flash Pricing on Quanta: 8 Concurrent Requests

01 · How many parallel coding agents can run on Quanta at once?

Eight requests can be in flight per Quanta account at the same time, and each key is limited to 120 requests per minute. Those are published and enforced figures, not suggestions buried in a help article. For a team running several coding agents against the same repository, those two numbers are the practical starting point for planning a parallel session.

Quanta is an API relay built for teams and heavy, sustained workloads. Concurrency, rate limits and invoices are spelled out instead of hidden, so an engineer can size a batch of agent runs against a defined ceiling rather than discovering the boundary through failed requests. The account-wide limit is 8 requests in flight; the per-key limit is 120 requests per minute. Each request body can be up to 25 MB, and an account can hold up to 20 live keys.

The practical reading is straightforward. A team running four agents has room to start another two without crossing the account ceiling. A team running eight agents is already at the ceiling and should think about how it schedules work. A team running twelve agents needs to queue or stagger, not simply launch everything at once.

Example: say your team runs six coding agents during a refactor. On Quanta, six of the eight in-flight slots are occupied, leaving two available for a quick verification pass or a focused sub-task. That is an example of how the published limit can shape a real workflow, not a promise about any particular workload.

02 · How does Gemini 3.8 Flash pricing work for agent traffic?

Gemini 3.8 Flash on Quanta is priced per 1M tokens, with input, output, cache writes and cache reads charged separately. The listed figures are $1.35 for input, $6.10 for output, $1.35 for cache writes and $0.1347 for cache reads. These are approximate USD figures; charges are made in CNY.

Quanta is pay-as-you-go. There is no monthly fee, no minimum spend and no subscription tier. You fund a prepaid balance, and each call is metered on real tokens. The dashboard lists the tokens and the charge for every call, which matters when a batch of agents is generating large volumes of output in a short window.

Because output tokens carry a higher listed rate than input tokens on Gemini 3.8 Flash, a team that encourages agents to print long transcripts will see a different cost profile from one that keeps prompts tight and results focused. The per-call breakdown in the dashboard lets you check that assumption directly instead of estimating from an aggregate.

For comparison within the Quanta catalogue, Gemini 2.5 Flash is listed at $0.2361 for input, $1.89 for output, $1.35... more precisely, $0.2361 for cache writes and $0.0236 for cache reads per 1M tokens. A worked example from the Quanta pricing sheet: at a 3:1 input to output ratio with no cache hits, Gemini 2.5 Flash costs approximately $6.49 for 10 million tokens a month and approximately $64.93 for 100 million tokens a month. These are examples, not quotes for a specific project.

Gemini 3.8 Flash is available through the native protocol only, through Gemini CLI. It is not presented as OpenAI-compatible, and Quanta does not describe it that way. One key and one balance work across the OpenAI-compatible, Anthropic native and Gemini native protocols, so a team can point Gemini CLI at Quanta without introducing a second billing relationship for the rest of its stack.

03 · Which clients pair with which models when several agents are running?

Client pairing on Quanta is defined by protocol, and it is worth setting out plainly before a team spreads a coding session across several tools.

• › Gemini models use the native protocol only, through Gemini CLI. That includes Gemini 3.8 Flash, Gemini 3.1 Pro, Gemini 4 Argon and Gemini 2.5 Flash.
• › Claude models work only from the official Claude Code and Claude Desktop clients, using Claude Code 2.1.285 or newer. Claude models do not work in Cursor, Cherry Studio, Cline, OpenCode, Codex, an SDK or cURL.
• › Codex, Cursor, Cherry Studio, Cline, OpenCode, OpenAI SDKs and cURL pair with GPT, DeepSeek and the other non-Claude, non-Gemini models in the Quanta catalogue, such as Qwen3.7 Plus, Qwen3.8 Max, Kimi K3, Hunyuan 4, Grok 4.6, Grok 4.7 and GLM-5.3.

This matters for parallel agent runs because a mixed client setup can otherwise produce confusing errors. If a team wants Gemini 3.8 Flash driving its agents, the route is Gemini CLI over the native protocol. If the same team also runs a coding session in Cursor or Codex, those sessions draw on the non-Claude, non-Gemini models, and they still draw on the same Quanta balance and the same published limits.

Streams also have a defined behaviour worth noting for long agent tasks. There is no total-time cap on a stream. A stream ends only after 300 seconds of upstream silence, so a slow but active coding run is not cut off by a fixed wall-clock limit.

04 · How does prepaid billing hold up when a large agent batch runs?

Quanta uses a prepaid balance with no monthly fee and no minimum spend. Top-up amounts are $1.39, $6.94, $13.89, $27.78, $69.44 and $138.89. A team can start small, watch the dashboard and add balance as usage becomes clear. Sign-up includes $0.50 in trial credit, which is enough to confirm that the client pairing and the protocol behave as expected before committing to a larger top-up.

Because every call is metered on real tokens and itemised in the dashboard, a large batch does not require a guess about what a parallel session will consume. The dashboard shows the tokens and the charge of each call, including the split between input, output, cache writes and cache reads. For a team that wants a wider view of model cost, Quanta also lists per-call rates for the rest of the catalogue. Claude Opus 5.5 is listed at $1.79 input, $8.46 output, $2.11 cache write and $0.0819 cache read per 1M tokens. A worked example: approximately $34.58 for 10 million tokens a month and approximately $345.83 for 100 million tokens a month, at a 3:1 input to output ratio with no cache hits. GPT-6.1 Sol is listed at $0.2083 input and $1.01 output per 1M tokens, with a worked example of approximately $4.10 for 10 million tokens a month and approximately $40.97 for 100 million tokens a month on the same basis. These are examples.

Unused balance can be refunded within 15 days of the top-up, which reduces the risk of funding a speculative workload. For company payment, bank transfer from a company account is supported, and invoices and advance quotes are arranged through the Contact page or the Messages page of the dashboard.

05 · What should a team configure before opening a Quanta account?

Start with the concurrency ceiling, because it is the constraint that determines how many agents can run in parallel. Eight requests in flight per account is the account-wide figure. The 120 requests per minute per key limit is separate, so distributing traffic across keys changes the request rate but not the number of requests the account can hold open at once. Up to 20 live keys are available per account, and each request body can be up to 25 MB.

Next, map each agent to the client and protocol that Quanta supports for its model family. Gemini CLI for Gemini models. Claude Code or Claude Desktop, at 2.1.285 or newer, for Claude models. The listed third-party coding clients for GPT, DeepSeek, Qwen, Kimi, Hunyuan, Grok and GLM models.

Then fund the balance. Choose a top-up that matches the size of the first batch, use the $0.50 trial credit to validate the setup, and watch the per-call token and charge breakdown as agents work. If the run needs more capacity, top up again rather than increasing the number of in-flight requests beyond 8.

Finally, note the data handling terms. The text users send is kept for 30 days for abuse handling and billing disputes, then deleted. Model answers are not stored. Quanta is an API relay for teams that need a clear account of what runs, how much it costs and where the limits sit, and opening an account at Quanta is a small step that puts all of those figures in one dashboard.