OpenAI tables suffer from small files and skew

Data engineering3 posts from 3 people3d active+11 post in the last 7 days, 1 the 7 days before (steady)

Posts per day3 posts, Sep 2 to Oct 21

The posts behind it

3, newest first

GitHub issuesOct 15 feat: send encrypted reasoning items back to the same model on stateless Responses API connections### Before Submitting - [x] Searched open/closed issues and discussions. Related #28240 covers sending model-bound reasoning to a **different** model; this report is loss of native `encrypted_content` on **same-model stateless Responses replay**. - [x] Reproduced with the...open-webui/open-webui · Kwentin32 repliesOpen ↗GitHub issuesOct 8 issue: streamed `usage` is summed per chunk, inflating `input_tokens` ~30x for providers that send cumulative usage on every chunk (Google OpenAI-compatible endpoint)### Before Submitting - [x] I searched open and closed issues and discussions for an existing report. (#27031 and #28528 are related but cover tool-loop and llama.cpp `timings` summing, not per-chunk summing.) - [x] I reproduced this bug on the latest release (v0.11.4,...open-webui/open-webui · lonelycodes2 repliesOpen ↗GitHub issuesSep 2 AI Assistant: mid-run observation compaction (#37234) sends two system messages, breaking strict OpenAI-compatible backends (System message must be at the beginning)...request:` prefix wraps the Jinja `raise_exception`). This makes the compaction feature unusable on a large class of self-hosted models, even though the compaction itself completes successfully. ### To Reproduce 1. n8n 2.38.1 (docker), AI Assistant with a custom OpenAI-compatible...n8n-io/n8n · chrisqianz4 repliesOpen ↗

Companies and products named

Named3 posts

OpenAI Named2 posts

llama.cpp Named1 post

Anthropic Named1 post

Qwen Named1 post

vLLM Named1 post

SQLite

About this problem

Evidence

3 posts from 3 people in 2 places, about 1 a week over 44 days. Mostly on open-webui/open-webui, n8n-io/n8n. Tools named alongside: llama.cpp, Anthropic, Qwen, vLLM.

Frustration Frustration 2 of 3· Seen on GitHub issues

How it was grouped

Posts that state a pain and match the "file-formats" rule about OpenAI. First post Sep 2, 2026, latest Oct 15, 2026. Corroborated: 3 or more posts from 2 or more people or places. Method

The same problem elsewhere

Elsewhere12 postsSmall files, skew and partitioning hurt performance Elsewhere6 postsClaude Code tables suffer from small files and skew Elsewhere2 postsModel Context Protocol tables suffer from small files and skew Elsewhere2 postsVercel tables suffer from small files and skew Elsewhere2 postsApache Flink tables suffer from small files and skew Elsewhere2 postsGitHub tables suffer from small files and skew Elsewhere2 postsChatGPT tables suffer from small files and skew Elsewhere2 postsClaude tables suffer from small files and skew

Other OpenAI problems

Same product3 postsFine-tuning OpenAI is costly and unpredictable Same product50 postsOpenAI users keep asking for help Same product8 postsRunning OpenAI locally hits memory and speed limits Same product15 postsOpenAI upgrades break working setups Same product9 postsTrouble with OpenAI chat Same product7 postsTrouble with OpenAI responses Same product5 postsOpenAI fails to return valid structured output Same product5 postsTrouble with OpenAI astra Same product4 postsTrouble with OpenAI custom Same product4 postsTrouble with OpenAI image Same product3 postsTrouble with OpenAI crashes Same product3 postsTrouble with OpenAI calls

History

CorroborationOct 18, 2026unverified to corroborated

MomentumOct 18, 2026fading to steady

MomentumOct 15, 2026steady to fading

AddedOct 9, 2026OpenAI tables suffer from small files and skew (2 posts)