Parasail
Open-model inference: serverless per token, dedicated per GPU-hour
Overview
Parasail runs inference on open models like DeepSeek, Qwen, GLM and Kimi, or your own weights, across a global GPU pool: serverless billed per token, Elastic private endpoints at 1.25x serverless rates, dedicated GPUs per hour, and batch at half price. Pay-as-you-go with no contracts.
Pricing
Pricing shown for reference only. These figures reflect RECATOOLS research as of 1 Oct 2026 and may be out of date or incomplete. This is not financial or purchasing advice — always confirm the current price on the provider’s official website before making any decision.
- No minimums, no rate limits
- DeepSeek, Qwen, GLM, Kimi and more
- Cached tokens 50% off
- Private endpoints
- Per-token billing
- B300, B200, H200, H100, RTX PRO 6000, RTX 5090
- Custom and fine-tuned weights
- Discount table by model size and precision
- Additional 50% off cached tokens
What you can produce with Parasail
- Serverless pay-per-token inference on open models
- Elastic private endpoints at 1.25x serverless rates
- Dedicated GPU deployments billed per GPU-hour
- Batch processing at 50% off serverless, cached tokens 50% off
- Support for custom/fine-tuned model weights
- SOC 2 Type II examination completed
Parasail covers most shapes a production inference workload takes: serverless per-token pricing with no rate limits, Elastic endpoints for private capacity at 1.25 times serverless rates, dedicated GPUs billed by the hour, and batch at half price with cached tokens halved again. Serverless prices are competitive, from a few cents per million tokens on small models. The company is young and moving fast — a $32 million Series A in April 2026, a SOC 2 Type II report, customers such as Elicit and Venice — but its boldest numbers, a fleet bigger than Oracle's cloud and 95% utilisation, are its own and unaudited. Note that dedicated capacity is now priced per GPU-hour, not per token as it once was. A strong option for teams serving open models who want flexibility without contracts; pilot on your traffic before committing.
What we found
Parasail sells inference on open models — DeepSeek, Qwen, GLM, Kimi, gpt-oss and others, or your own weights — on GPUs it sources from providers around the world. It launched in April 2025 claiming a fleet larger than Oracle's entire cloud, a claim TechCrunch attributed to Parasail rather than verified, and in April 2026 raised a $32 million Series A co-led by Touring Capital and Kindred Ventures with Samsung NEXT, Flume and Banyan, for $42 million in total. It says it was processing 500 billion tokens a day at that point and growing revenue 30% a month. Its co-founders are chief executive Mike Henry, formerly chief product officer at Groq and founder of Mythic, and Tim Harris, founder of Swift Navigation. Named customers include Elicit, mem0, Kotoba and Venice, and it has completed a SOC 2 Type II examination.
Its product shapes have changed since earlier descriptions, including ours. Serverless inference bills per million tokens with no minimums or rate limits, from $0.04 input and $0.20 output for gpt-oss-20b to $3 and $15 for Kimi K3; DeepSeek V4 Flash costs $0.14 and $0.28. Dedicated deployments are no longer billed per token: they are priced per GPU-hour, billed by the minute, on B300, B200, H200, H100, RTX Pro 6000 and RTX 5090 hardware. Customers who want private capacity priced by the token now buy Elastic endpoints, set up through sales for selected models at 1.25 times the serverless rate. Batch jobs cost half the serverless price, and cached tokens take a further 50% off.
Parasail attributes its economics to high fleet utilisation — around 95%, according to its own blog — which, like its token volume, has not been independently audited. Billing runs in arrears, with the card charged once usage reaches $25, and no free credits are published. The early customer praise most often quoted, from Elicit's chief executive, comes from Parasail's own launch announcement. Pilot it on your own traffic before relying on its claimed capacity.
Compiled by RECATOOLS from public sources: vendor documentation and pricing, code repositories and published reporting.
About this listing
This entry was compiled from publicly available data including Parasail's official website, press releases, documentation, and reputable third-party publications. RECATOOLS is not affiliated with Parasail unless explicitly stated.
Third-party AI tools update their pricing, features, availability, and policies frequently. Information here may be outdated by the time you read this — we make reasonable efforts to keep listings current, but cannot guarantee absolute accuracy.
For the latest details, please refer to Parasail directly →
Spotted something out of date? Suggest an update →
Alternatives to Parasail
More in LLMs & Chat
