LLM APIs still disagree on the parameter name for output token limits
nuno6Varnish · reddit · 2026-07-29
LLM APIs still disagree on the name for output token limits
A Reddit post points out a common source of API failures: different LLM providers expect different parameter names for maximum output tokens.
Examples mentioned
- OpenAI uses maxcompletiontokens
- Anthropic uses maxtokens
- Qwen uses extrabody.enablethinking instead of a standard thinking.type
- Setting topp, topk, or temperature can also trigger 400 errors on some flagship models
The post argues these mismatches happen millions of times a day and asks what other provider-specific quirks people keep running into.
More from coding & agent
- Relume MCP and Claude Code power a Markdown-first design workflow — michalmalewicz · 2026-07-29
- A research talk compares SWE-bench, CodeClash, and ProgramBench for agentic coding — OfirPress · 2026-07-29
- Kuna debuts as an agent-driven Rust decompiler, ranking near Hex-Rays in structuring — moyix · 2026-07-29
- $100k Bet: AI 'Vibecoding' Can't Match Complex SEO Tools Like Ahrefs — eptwts · 2026-07-29
- An AI-agent startup argues billing should work from facts, not scattered API calls — davidcrawshaw · 2026-07-29
- Most Cursor users still treat it like VS Code, despite 1M users and a $29.3B valuation — nikola_mr64990 · 2026-07-29