This is a revised edition of an earlier document, verified against and updated to reflect each platform's actual free-tier policy as of July 2026. Corrections and additions relative to the original are listed separately at the bottom of this document.
1. Summary: The Reality of Free Tiers in 2026
Since 2022, permanent free tiers have been eliminated one after another across the PaaS industry.
| Year | Event |
|---|---|
| November 2022 | Heroku ends its free plan (cheapest paid tier: Eco, $5/month) |
| August 2023 | Railway eliminates its permanent free tier, switching to a one-time $5 trial model |
| October 2024 | Fly.io eliminates free resource allocation for new signups |
| September 2025 | Render shortens the free-service sleep timeout (30 minutes -> 15 minutes) |
As a result, as of 2026, Render and PythonAnywhere are essentially the only options for running a Python web service on a genuinely permanent free tier; everything else is either trial credit or conditionally free.
2. Platform Comparison Table
| Platform | 2026 Free Tier Status | Pros | Cons / Limitations |
|---|---|---|---|
| Render | Permanent free tier maintained. 750 instance-hours/month per workspace, 512MB RAM, 0.1 vCPU, 100GB/month bandwidth, 500 build minutes/month. No card required | Manage a web server + PostgreSQL + Redis (Key Value) + Cron all in one place. Automatic deployment via Git integration. The most solid Heroku alternative among free options | Sleeps after 15 minutes of no traffic, with a 30-60 second cold start.Free PostgreSQL is capped at 1GB and expires 30 days after creation (data is deleted if not upgraded within 14 days of expiry). Using self-pinging to avoid sleep may violate the terms of service |
| PythonAnywhere | Permanent free tier maintained. 512MB disk, 1 web app (username.pythonanywhere.com) |
Provides a web IDE for coding and deploying directly in the browser. Optimized for Django/Flask (WSGI). The lowest barrier to entry for beginners | External network access is restricted to a whitelist of domains. No custom domains. ASGI (e.g., FastAPI) support is in beta and limited |
| Vercel | Hobby plan is free (butrestricted to personal, non-commercial use). Fluid Compute applied by default, billed on Active CPU (within the free quota) | Officially supports zero-config FastAPI deployment (since September 2025). An Active CPU billing model that doesn't charge for I/O wait time. Top-tier developer experience, including preview deployments. Strong for Next.js frontend + Python API combinations | Function execution time is limited (requires setting maxDuration). Not well suited to persistent connections like WebSockets. Commercial services violate the Hobby plan's terms |
| Fly.io | No free tier for new signups. A small trial credit followed by pay-as-you-go billing. Only legacy-plan accounts retain their prior free allocation (e.g., 3 shared VMs) | Deploys containers across 30+ regions for low global latency. Supports WebSockets and persistent connections. The smallest VM is as cheap as around $2/month | Not free. Requires DevOps skills (Dockerfile, etc.). Metered egress billing (e.g., $0.04/GB in Asia) makes costs hard to predict. Leaving the legacy plan means you can't go back |
| Railway | No permanent free tier. A one-time $5 trial credit (30 days) upon signup. After that, a minimum Hobby plan of $5/month plus metered usage | Ultra-fast, template-based provisioning. Automatic framework detection, deploy via Git push. Simple database provisioning | Mandatory conversion to paid after the trial is used up. Per-second metered billing frequently results in bills larger than expected. Not well suited for advanced network configuration or compliance requirements |
| Heroku | No free plan (ended November 2022). Cheapest tier is Eco at $5/month (1,000 dyno-hours, with sleep) | A mature ecosystem, extensive add-ons and documentation. Proven stability | Even the cheapest plan requires payment information. Excluded from the free-tier comparison |
3. Additional Options Worth Considering (Not in the Original)
| Platform | Free Tier | Best Fit |
|---|---|---|
| Google Cloud Run | 2 million requests/month, free vCPU/memory allocation (requires a card on file) | Container-based Python APIs. Scale-to-zero makes it effectively free for small-scale SaaS |
| Koyeb | Provides a small free instance | A Render-like PaaS, centered on European regions |
| Cloudflare Workers | 100,000 requests/day free, includes free D1 (SQLite) and R2 allocation | Python Workers are in beta, so maturity should be treated with caution. Well suited to edge APIs |
| Hugging Face Spaces | Free CPU instances | Gradio/Streamlit-based demos, ML prototypes |
| Oracle Cloud Always Free | Permanently free ARM VM (4 OCPU/24GB) | Effectively a free VPS. Carries the highest operational burden of self-management |
4. Selection Guide
- The easiest, fastest way to a full-stack MVP: Render. Just be sure to account for the 30-day expiry of the free PostgreSQL database - if your data matters, either start with Starter ($7/month) or higher, or pair it with an external free database (Neon, Supabase, etc.) from the beginning.
- For learning Python / education: PythonAnywhere. Still the best option for learning Django/Flask. Not a good fit if you're centered on FastAPI.
- Next.js frontend + Python API: Vercel. Zero-config FastAPI support has significantly improved Python backend suitability compared to when the original document was written. However, the non-commercial-use restriction and inability to hold persistent connections still apply.
- Global low latency / WebSockets: Fly.io. That said, it should now be classified not as a "free alternative" but as a "cheap paid alternative," and should be considered with a budget of roughly $5-20/month in mind.
- Fast prototyping with an expected eventual switch to paid: Railway. A natural flow of validating with the $5 trial, then transitioning to Hobby.
- Zero-cost, always-on operation as the top priority: Google Cloud Run (scale-to-zero) or an Oracle Always Free VM.
5. Corrections Relative to the Original Document
| # | Original claim | Correction | Severity |
|---|---|---|---|
| 1 | Railway: "efficient because you only pay for what you use within the free tier" | The permanent free tier was eliminated in August 2023. Currently there is only a one-time $5 trial (30 days), after which a minimum Hobby plan of $5/month applies. Describing it as a "free tier" at all is inaccurate | High |
| 2 | Fly.io: "offers a generous free tier" | All free allocation for new signups was eliminated in October 2024. Only legacy-plan accounts retain prior benefits. New users get a small credit followed by fully metered billing | High |
| 3 | Render: "may go to sleep if unused for a period of time" | This is a confirmed spec, not an estimate. Sleeps after 15 minutes of no traffic (shortened from 30 minutes in September 2025), 30-60 second cold start, 750-hour monthly cap | Medium |
| 4 | Render's cons section omitted the free database's expiry | Free PostgreSQL expires 30 days after creation. From an operational standpoint this is a more critical constraint than sleep behavior, so it must be included | High |
| 5 | PythonAnywhere: "500MB of storage" | 512MB is the precise figure. Also, "weak on async (ASGI)" remains directionally valid, but the start of beta ASGI support should be reflected | Low |
| 6 | Vercel: "not suited for long-running Python jobs" | The direction is still valid but the description is outdated. With the introduction of Fluid Compute (2025), zero-config FastAPI support, Active CPU billing (no charge for I/O wait), and an adjustable maxDuration are now available. WebSocket limitations and the Hobby plan's non-commercial-use restriction remain valid, so these should be listed as the core cons instead |
Medium |
| 7 | Heroku: "free plan discontinued starting 2022" | More precisely, it ended on November 28, 2022. Noting that the cheapest alternative is Eco at $5/month completes the comparative context | Low |
| 8 | Overall: three-way free-tier classification ("permanent free / generous free / discontinued") | This classification itself has collapsed as of 2026. The actual landscape is: "permanently free (Render, PythonAnywhere) / conditionally free (Vercel non-commercial) / trial only (Railway, Fly.io) / no free tier (Heroku)" | High |
6. Additional Suggestions Relative to the Original Document
- Add an industry-trends section: explicitly noting the free-tier elimination trend running from Heroku -> Railway -> Fly.io provides context for "why the options are so narrow right now."
- Add alternative platforms: Google Cloud Run, Koyeb, Cloudflare Workers, Hugging Face Spaces, Oracle Always Free. Cloud Run in particular deserves substantial coverage as a practically free way to operate a small-scale Python SaaS.
- Cover database strategy separately: since compute and database free policies move independently (e.g., Render's free compute vs. its 30-day database expiry), the pattern of pairing with free managed PostgreSQL providers like Neon/Supabase deserves its own section.
- A caution note on sleep avoidance: the widely shared technique of self-pinging with tools like UptimeRobot to bypass sleep should come with a note that Render may treat this as abnormal traffic and suspend the account.
- State the verification date explicitly: since free-tier policies shift quarter to quarter, the document should clearly state an "as-of" date at the top, and attach links to the official pricing pages as references.
References
- Render official docs (Deploy for Free): https://render.com/docs/free
- Render Pricing: https://render.com/pricing
- Vercel FastAPI docs: https://vercel.com/docs/frameworks/backend/fastapi
- Vercel Fluid Compute: https://vercel.com/docs/fluid-compute
- Railway Pricing: https://railway.com/pricing
- Fly.io Pricing: https://fly.io/docs/about/pricing/
- PythonAnywhere Plans: https://www.pythonanywhere.com/pricing/
As of: July 19, 2026. Free-tier policies change frequently, so re-checking the official pages before deployment is recommended.