Ollama Latency Blowup

·2026.04.15 23:46

Key point

Complaints about Ollama Cloud's response latency and low TPS have escalated to the point of subscription cancellations.

Details

A complaint has been raised that Ollama Cloud's inference speed has become unusable in practice.

When an actual user ran several models, the token generation speeds per second were as follows.

GLM 5: 11 tok/s
GLM 5.1: 8 tok/s
Qwen 3.5: 14 tok/s
MiniMax 2.7: 22 tok/s

The author claims that even simple tasks take over 1 hour, and questions why they're paying the subscription fee. They pointed out that the drop in inference speed is a bigger problem than the reduced usage limits, and stated they would cancel their subscription if the complaint isn't resolved.

ollama llm inference latency benchmark cloud subscriptions

Read originalShare on Threads

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.