Llama 3.1 8B Instruct (FP8) on Cloudflare
Cloudflare Workers AI Free API tier
The catch: 10,000 Neurons a day is small (about 0.37M output tokens on gpt-oss-20b, about 49K on Llama 3.3 70B), and on Workers Free calls fail until 00:00 UTC once it is spent. The newest large models (Kimi, GLM-5.x, DeepSeek V4) need Workers Paid ($5 a month minimum) or prepaid credits.
| Access | Free API tier |
|---|---|
| Free limits | Uses the account's shared 10,000 free Neurons a day (reset 00:00 UTC) and costs 13,778 neurons per M input tokens, 26,128 neurons per M output tokens, so the free pool covers roughly 383K output tokens a day on this model alone (derived); text generation is limited to 300 requests/min. |
| Modality | text |
| Credit card | Not required |
| Commercial use | Allowed |
| Trains on your data | No |
| Licence or terms | Cloudflare Self-Serve Subscription Agreement (plus each model's own licence) |
| Context window | 32,000 tokens |
| Model id | @cf/meta/llama-3.1-8b-instruct-fp8 |
| Base URL | https://api.cloudflare.com/client/v4/accounts/{account_id}/ai/v1 |
| Last checked | 8 Oct 2026 |
How to use Llama 3.1 8B Instruct (FP8)
It speaks the OpenAI API at https://api.cloudflare.com/client/v4/accounts/{account_id}/ai/v1, so the OpenAI SDK you already use works once you change the base URL and key.
curl https://api.cloudflare.com/client/v4/accounts/$ACCOUNT_ID/ai/v1/chat/completions -H "Authorization: Bearer $CLOUDFLARE_API_TOKEN" -H "Content-Type: application/json" -d '{"model":"@cf/meta/llama-3.1-8b-instruct-fp8","messages":[{"role":"user","content":"Hello"}]}'I want to use the free AI model "Llama 3.1 8B Instruct (FP8)" from Cloudflare (model id: @cf/meta/llama-3.1-8b-instruct-fp8) in my project. Please: 1. Check the provider docs and give me the exact steps to get a free API key. 2. Give me a minimal working example that reads the key from an environment variable. 3. Tell me the real free limits and any catch (commercial use, card, data use). Listed limits: 10K neurons/day · 13,778 in / 26,128 out per M tokens. 4. If the same provider has a more generous free option, mention it.
Before you ship on it
- 10,000 Neurons a day is small (about 0.37M output tokens on gpt-oss-20b, about 49K on Llama 3.3 70B), and on Workers Free calls fail until 00:00 UTC once it is spent. The newest large models (Kimi, GLM-5.x, DeepSeek V4) need Workers Paid ($5 a month minimum) or prepaid credits.
- The Self-Serve Subscription Agreement (last updated 12 Sep 2025) has no non-commercial clause; free services run until Cloudflare ends them 'in our sole discretion'. The data-usage page says each model may carry its own…
- Free tiers are shared capacity: treat the limits as a ceiling, not a promise, and keep a paid fallback.
Questions
Is Llama 3.1 8B Instruct (FP8) on Cloudflare free?
Yes. Cloudflare offers it as free api tier, with these limits: 10K neurons/day · 13,778 in / 26,128 out per M tokens. No credit card is needed.
Can I use Llama 3.1 8B Instruct (FP8) on Cloudflare commercially?
Yes, under Cloudflare Self-Serve Subscription Agreement (plus each model's own licence). Check the current terms before you ship.
How do I start with Llama 3.1 8B Instruct (FP8) on Cloudflare?
It speaks the OpenAI API at https://api.cloudflare.com/client/v4/accounts/{account_id}/ai/v1, so the OpenAI SDK you already use works once you change the base URL and key.
Source: developers.cloudflare.com/workers-ai/models/llama-3.1-8b-instruct-fp8/, checked 8 Oct 2026.