The latest articles on DEV Community by Imad ERRACHIDI (@imad_errachidi_0c00845672). https://dev.to/imad_errachidi_0c00845672 https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F4225251%2F373332a7-89ee-406b-b8f9-fbfa68eb7306.png https://dev.to/imad_errachidi_0c00845672 en Imad ERRACHIDI Wed, 04 Nov 2026 02:24:55 +0000 https://dev.to/imad_errachidi_0c00845672/i-track-how-much-each-provider-charges-for-the-same-open-llm-llama-33-70b-ranges-from-0155-to-9me https://dev.to/imad_errachidi_0c00845672/i-track-how-much-each-provider-charges-for-the-same-open-llm-llama-33-70b-ranges-from-0155-to-9me <p><em>Disclosure: this is my own project.</em></p> <p>If you call an open-weight model through an API, the price you pay depends heavily on <strong>who</strong> serves it. The same model, same weights, can cost several times more at one provider than at another.</p> <p>I built a small monitor that checks this every 5 minutes. Here's how it works and what it's finding.</p> <h2> How it works </h2> <ul> <li>Every 5 minutes it reads OpenRouter's <strong>public</strong> per-model endpoint data (no API key needed). That list includes every provider serving a model and its per-token price.</li> <li>For each provider it computes a <strong>blended cost per 1M tokens</strong>, assuming 3 input tokens for every output token.</li> <li>It excludes endpoints that are currently down (negative status or 0% recent uptime).</li> <li>It computes the median, mean and standard deviation per model, and flags providers that are <strong>15% or more below the median</strong>.</li> <li>It warns when the cheapest provider uses a different <strong>quantization</strong> than the most expensive one, because fp4 vs fp16 is not a fair comparison.</li> </ul> <p>It runs as an n8n workflow with a duplicate filter, so the same provider/model pair is not re-alerted for 6 hours unless the price moves by more than 5%.</p> <h2> What it's finding right now </h2> <div class="table-wrapper-paragraph"><table> <thead> <tr> <th>Model</th> <th>Cheapest</th> <th>Median</th> <th>Providers</th> <th>Note</th> </tr> </thead> <tbody> <tr> <td>Llama 3.3 70B</td> <td>DeepInfra $0.155/M</td> <td>$0.64/M</td> <td>9</td> <td>fp8 on both ends</td> </tr> <tr> <td>Llama 3.1 8B</td> <td>DeepInfra $0.025/M</td> <td>$0.12/M</td> <td>4</td> <td>quantization differs</td> </tr> <tr> <td>Gemma 3 27B</td> <td>DeepInfra $0.10/M</td> <td>$0.149/M</td> <td>4</td> <td>fp8 on both ends</td> </tr> <tr> <td>gpt-oss-120b</td> <td>~$0.04-0.06/M</td> <td>$0.20/M</td> <td>24</td> <td>cheapest are fp4/int4</td> </tr> </tbody> </table></div> <h2> When this is (and isn't) useful </h2> <p>If you route through OpenRouter's default settings, you are probably already getting the cheap end, since its default routing favours price.</p> <p>It's more useful if:</p> <ul> <li>you're contracted <strong>directly</strong> with one provider and never compared,</li> <li>you pin providers manually for latency, context length or quantization and want to know what that choice costs,</li> <li>you're choosing where to self-serve volume.</li> </ul> <h2> Caveats </h2> <p>These are <strong>list prices</strong>. They ignore latency, throughput, rate limits and reliability, and a cheaper endpoint may have a shorter context window or lower precision. Check those, and the provider's terms, before switching.</p> <h2> Follow the alerts </h2> <p>Alerts are posted to a free Telegram channel: <a href="https://t.me/+CWPuVJAQrXozYTZk" rel="noopener noreferrer">https://t.me/+CWPuVJAQrXozYTZk</a></p> <p>Feedback on the methodology is very welcome, especially on better ways to weight input vs output tokens or to factor in throughput.</p> ai llm showdev automation