CareerPocket AI is built so you bring your own AI — no subscription, no company holding your data hostage. Most of the time, that means Google's Gemini, which offers a genuinely generous free tier. But every so often, people hit an error, assume the app is broken, and give up right when they needed it most.
It's not a bug in CareerPocket. It's a real, well-documented pattern with brand-new free AI models — and once you understand it, it's easy to work around.
Worth knowing upfront: paid AI models — whether from Google, Anthropic (Claude), or OpenAI — are generally the most stable option, since providers reserve dedicated capacity for paying customers. Free-tier access is a genuinely different arrangement: Google offers it partly so a large number of real users try each new model quickly, which helps them find issues and tune performance before wider rollout. That's great for you as a free user in terms of access to cutting-edge AI at no cost — but it also means the newest free models attract a rush of people testing them all at once, right when Google's own capacity for that specific model is still ramping up. That combination is exactly what tends to cause temporary "unavailable" errors.
The short version
Google regularly ships new, more capable versions of its free Gemini models — approximately every few weeks recently. Every time a new one launches, it takes Google's own infrastructure a while to catch up with demand. In the meantime, that specific model can return "temporarily unavailable" errors more often than usual — not because your account or your request did anything wrong, but because the model itself is genuinely overloaded on Google's end.
This has happened with essentially every recent Gemini Flash release. It's a real cost of getting access to the newest AI capability for free, not a CareerPocket problem.
Why we don't just pick "the stable one" for you
We could quietly default everyone to an older, more established model and call it a day. We don't, for one honest reason: it's genuinely a tradeoff, not a clear winner, and it's your account, your free-tier usage, your choice to make — not ours to make for you silently.
The newest models are usually meaningfully better at following complex instructions — which matters for something like tailoring a resume to a specific job. The older, established models are far more reliable and let you do far more per day for free. Neither is simply "the right answer" for everyone.
Your 4 practical options right now
| Model | Released | Best for | Daily free limit* | Trade-off |
|---|---|---|---|---|
| Gemini 3.8 Flash (newest) | Sep 2, 2026 | Best quality, most capable | ~20 requests/day | Can be unstable in its first few weeks |
| Gemini 3.7 Flash | Aug 13, 2026 | Near-newest quality | ~20 requests/day | Same new-model instability risk |
| Gemini 3 Flash (Preview) | Dec 17, 2025 | Strong quality, more established | ~20 requests/day | Still limited daily volume |
| Gemini 3.1 Flash Lite | May 7, 2026 | Heavy daily use, most reliable | ~500 requests/day | Solid but slightly less polished writing |
*Free-tier limits are set by Google and can change without notice — always double-check current limits in your own Google AI Studio account if it matters for your usage.
What CareerPocket does about it automatically
If you're on one of the two newest models (3.7 or 3.8 Flash) and a generation fails, you'll now see two buttons right in the error message — not a trip to Settings, not guesswork:
- Try Gemini 3.1 Flash Lite — for when you want to keep working right now, with far higher daily volume
- Try Gemini 3 Flash Preview — for when you want to stay closer to the newest quality
One click switches your model and retries the exact same thing you were doing — nothing you'd already written is lost.
How to check current free models and limits yourself
Everything in the table above reflects what's true as we publish this — but Google changes free-tier availability and limits without much notice (it's happened before). If you want to see exactly what's free on your own account right now, it takes about a minute:
- Go to Google AI Studio — aistudio.google.com — and sign in with the same Google account you use for your API key.
- Open the "Rate Limit" page in the left sidebar (under Usage & Billing). If you don't see it, click your account/dashboard icon first.
- Look at the "Free tier" label at the top of the page — this confirms you're viewing your actual current free-tier access, not a paid view.
- Scan the model list. Each model shows real numbers for RPM (requests per minute) and RPD (requests per day). A model showing real numbers (e.g. "5/5" or "15/500") is genuinely free and usable. A model showing "0/0" is not available on your free tier.
This is the single most reliable way to confirm what's actually free right now — more reliable than any article (including this one), since Google can change it at any time.
Our honest recommendation
If you're actively job hunting and generating multiple documents a day: start with Gemini 3.1 Flash Lite. The daily volume matters more in practice than the small quality difference, and you're far less likely to hit an error mid-application.
If you're sending a small number of applications and want the strongest possible writing: try 3.7 or 3.8 Flash first, and let the one-click fallback catch you if it's having a rough day.
Still stuck?
If switching models doesn't help, it's worth checking your own account's usage in Google AI Studio directly, or trying a completely different provider (OpenAI, Claude, or a local model) in Settings.