Replicate Review
Run open-source machine learning models with a cloud API
Is this your product? Claim this page · Request a change
Looking for a Replicate alternative? See our ranked comparison.→What is Replicate?
Replicate is a cloud API for running machine learning models without managing GPU infrastructure yourself. You call one of thousands of community-contributed models (image generation, speech, video, LLMs) with a single API request, or push your own model packaged with their open-source tool Cog and get an API endpoint for it.
Billing is per-second of compute for most models, or per-token or per-output-unit for some newer ones like Claude or Flux. Private model deployments run on dedicated hardware you control (instance type, min/max instances, scale-to-zero), which means you're billed for idle and boot time in addition to inference.
Cloudflare announced its acquisition of Replicate in November 2025 to fold it into Cloudflare Workers AI. As of mid-2026 Replicate still operates as its own product and pricing page, with Replicate's careers page redirecting to Cloudflare's.
Replicate screenshots
Who it's for
- ✓ Teams that want to call a huge catalog of open and third-party models through one API without standing up their own GPU fleet
- ✓ Bursty or spiky inference workloads where per-second billing beats paying for reserved capacity
- ✓ Engineers who already package models with Cog and want a fast path from a model repo to a production endpoint
Who should look elsewhere
- ✗ Teams running always-on, high-utilization inference at scale, where reserved or committed-use GPU pricing elsewhere is cheaper per hour
- ✗ Teams that need a hard monthly spend cap without manually managing prepaid credit, since Replicate dropped account spend limits in mid-2025
- ✗ Teams that need enterprise compliance guarantees baked in without negotiating a custom deal
Pros
- + Pricing is transparent and published down to the per-second hardware rate, with no plan tiers or seats to negotiate
- + Huge, actively growing catalog of public models you can call with no setup
- + Cog packaging makes it straightforward to move a model from a repo into a hosted API
- + Recent additions like agent skills and MCP server auto-discovery make it easier to wire Replicate into coding assistants and agent workflows
Cons
- – Per-second GPU rates run noticeably higher than some competitors for sustained workloads; an H100 on Replicate lists at $5.49/hr versus $3.95/hr on Modal and $3.99/hr list price ($1.89/hr at the low end) on Fal.ai for comparable hardware
- – Private deployments bill for the full instance lifecycle, so cold starts and idle warm instances add real cost beyond actual inference time
- – Monthly account spend limits were deprecated on 2025-07-01; the only remaining hard cap is a prepaid credit balance, which raises the risk of runaway bills if you rely on soft usage alerts
- – No permanent free tier; the free-to-try model list is limited to a curated set of models and the exact run allowance isn't disclosed
Replicate pricing
What you pay for
You pay for compute time or output units, not for seats. Most public models bill by the second of GPU or CPU time the run takes; some newer models bill per token or per output image/video instead. Pricing itself is fully disclosed on the public pricing page down to the per-second hardware rate, but your actual bill depends on run time, idle time on private deployments, and which billing method a given model uses, so it takes some math to predict.
You pay for what you consume rather than a per-seat fee, so cost scales with usage.
| Plan | Price | Highlights |
|---|---|---|
| Pricing is quote-only. Contact the vendor for a quote. | ||
No plans or seats. You pay per second of compute on the hardware you use, or per token/output unit for models that bill that way. CPU (small) starts at $0.000025/sec ($0.09/hr). Nvidia H100 runs $0.001525/sec ($5.49/hr). Language models like Claude 3.7 Sonnet bill per token ($3.00/million input tokens on Replicate). Image models like Flux 1.1 Pro bill per output image ($0.04/image). Private deployments are billed for the whole instance lifecycle (boot, idle, and active time), not just inference, so a slow cold start or an idle warm instance still costs money. Replicate removed the ability to set a monthly account spend limit on 2025-07-01; the only hard caps now are prepaid credit balances. There is no permanent free tier, only limited, undisclosed free runs on a curated set of models. Source: vendor pricing page fetched 2026-07-07.
Pricing verified July 7, 2026 · source
How Replicate's pricing compares
Replicate next to its closest alternatives on entry price, billing, and whether pricing is public.
| Tool | Starting price | Billing | Free option | Pricing disclosed |
|---|---|---|---|---|
| Replicate | $0.000025/second | usage-based | No | Public |
| Baseten | Custom / quote | usage-based | No | Not disclosed |
| Modal | $0.0000131/core-second | usage-based | Yes | Partly public |
| Together AI | Usage-based | usage-based | No | Public |
| Fireworks AI | Usage-based | usage-based | Trial ($1 in free credits for new accounts to test serverless inference) | Partly public |
| RunPod | Usage-based | usage-based | No | Public |
Is Replicate still actively developed?
Last significant update: April 2026. Replicate published a set of agent skills (markdown instruction files following the open Agent Skills spec) that teach coding assistants like Claude Code how to discover, compare, and run models on the platform, including prompting techniques for image and video models.
Top Replicate alternatives
Replicate FAQ
Does Replicate have a free tier?+
No permanent free tier. A small, curated set of models offers limited free runs, but there's no ongoing free plan and no disclosed run allowance.
How is Replicate priced?+
Usage-based. Most models bill per second of the GPU or CPU time a run takes, at rates that vary by hardware (from about $0.000025/sec for small CPU to $0.001525/sec for an H100). Some newer models bill per input/output token or per output image instead.
Can I cap how much I spend on Replicate?+
Not with a monthly spend limit anymore. Replicate deprecated that feature on 2025-07-01. The remaining option is to run on prepaid credit, which stops work once the balance hits zero, or to monitor usage manually and disable auto-reload.
Is Replicate still independent after the Cloudflare acquisition?+
Cloudflare announced its agreement to acquire Replicate in November 2025 to build it into Cloudflare's AI platform. Replicate's product and pricing page were still operating under the Replicate brand as of mid-2026.