2026-08-14Local Claude parallel subagent recheck (read directly from replicate.com; still pay-as-you-go)
2026-08-07 21:55:132026-08-07 21:55 Checked: Replicate's pricing plans include public models and private models. Public models are charged based on output or execution seconds, while private models are charged based on the uptime of dedicated hardware. Prices range from US$0.025 per image to US$0.25 per second, with multiple models available. Source: https://replicate.com/pricing
2026-08-03 03:43:21The official pricing page was successfully read directly through the r.jina.ai agent, and the prices remain unchanged. Source: https://replicate.com/pricing
2026-08-03 03:34:37(官網 replicate.com/pricing 經r.jina.ai代理直讀) 所有硬體費率(CPU Small US$0.000025/sec~H100 US$0.001525/sec)及模型費率(FLUX Pro US$0.04、FLUX Dev US$0.025、Ideogram v3 US$0.09、Claude 3.7 US$3.00/M input)皆無變動，無新促銷。
2026-07-31Initial documentation. (Directly from the official website replicate.com/pricing) Since the free tier description is not seen, free_tier=false. The GPU per-second rate is between US$0.000025 for CPU Small and US$0.001525 for H100. The flagship models are priced as follows: FLUX Pro at US$0.04/image, FLUX Dev at US$0.025/image, Ideogram v3 Quality at US$0.09/image, and WAN 2.1 720p at US$0.25/second. Private models are charged based on the total uptime, including idle time.
Continuously re-checked, every fact dated
What Is This
Thousands of open-source generation models are available on demand, with pricing per second.
Free Version Limitations
The official pricing page does not offer a free tier or trial quota (Checked 2026-07-31 directly from replicate.com/pricing; whether registration comes with a gift quota is pending verification)
Free Usage Limit
Features
Free quota
Description
Free tier
None
The official website's pricing page does not provide information on the Free quota.
Minimum consumption
—
Pay-as-you-go, only charge for what you use
Quota Resets At
—
Terms Of Use
Requires registration and binding a payment method
Free Version For Commercial Use
Yes
Common Pitfalls
It **doesn't have a Free quota**, but is billed by the second, so you only pay for what you use, making it cheaper for occasional use than a monthly subscription — generating 100 FLUX Dev images costs around US$2.5. What you really need to pay attention to is "private models": dedicated hardware will **charge you for idle waiting time**, not just actual computation.
Varies by model: FLUX Pro US$0.04/image, FLUX Dev US$0.025/image, Ideogram v3 Quality US$0.09/image, WAN 2.1 video (720p) US$0.25/second, Claude 3.7 Sonnet US$3.00 per million input tokens (Checked on official website as of 2026-07-31)
Which Model
FLUX, Ideogram, WAN video models, and thousands of other open-source models
Usage Quota
Charged by the number of images or seconds, pay for what you use.
Max Reading Time
Based on model specifications
This Plan Includes
FLUX Dev US$0.025/image
FLUX Pro US$0.04/image
Ideogram v3 Quality US$0.09/image
WAN 2.1 Video 720p US$0.25/second
Free to use, no charge
Not Included
Free quota
Monthly subscription with unlimited generation
What Sets Us Apart
Compared to Midjourney (starting at US$10/month), if you generate less than 400 images per month, pay-per-image is more cost-effective here; however, it lacks the parameter tuning community and interface of Midjourney.
Only the actual processing time is calculated. GPU per-second rate: CPU Small US$0.000025 (US$0.09/hour), CPU US$0.0001 (US$0.36/hour), T4 US$0.000225 (US$0.81/hour), L40S US$0.000975 (US$3.51/hour), A100 80GB US$0.0014 (US$5.04/hour), H100 US$0.001525 (US$5.49/hour) (Checked on 2026-07-31 from the official website)
Which Model
Any public model, customizable GPU level
Usage Quota
Only count actual processing time
Max Reading Time
Based on selected hardware
This Plan Includes
T4 US$0.81/hour
L40S US$3.51/hour
A100 80GB US$5.04/hour
H100 US$5.49/hour
Not Included
Idle time is free (only for public models; private models are charged even when idle)
What Sets Us Apart
It saves the hassle of setting up the environment compared to renting a cloud GPU, but the unit price is higher. For long-term large-scale operations, self-deployment is still more cost-effective.
Dedicated instances are billed based on "total uptime", including startup and idle waiting time, not just actual computation time; fine-tuned models with fast startup are exceptions, only actual processing time is counted (Checked 2026-07-31 from official website)
Long-form writing and document quality are its strengths, with the paid version including the Claude Code engineering tool
FreeThe quota is calculated based on a rolling 5-hour usage window (not simply resetting at a fixed number every day), and general conversations can be used for around 10-20 times, excluding Claude Code.
Gemini's image generation model went viral for its consistency in character depiction and explosive editing capabilities
FreeThe free version of Gemini includes a small amount of Nano Banana generation quota; paid plans increase the quota limit based on the level (official website confirmed on 2026-07-30 that the official name is "Nano Banana Pro", with a tiered system of Plus = limited access, Pro = more access, and Ultra = higher or highest access, but the official website does not disclose the specific quota limit, only estimated by third-party sources, not officially confirmed).
Model aggregation platform: It integrates thousands of open-source models (FLUX, Stable Diffusion, video and voice models) into a unified API, eliminating the need to prepare a graphics card. Unlike OpenRouter, it focuses on generating media such as images, videos, and voices, while OpenRouter focuses on text-based dialogue models. The reason for marking free_tier as false: The official pricing page does not provide any information on a free tier, and if there is a quota gift upon registration, it needs to be checked and corrected separately.
