svg.reasoning.effort values you can send depends on svg.model: mainstream author models such as gpt-5.6-terra and gpt-6-luna accept none / low / medium / high / xhigh / max, and minimal returns 400 — it belongs to the original GPT-5 generation (gpt-5 / gpt-5-mini / gpt-5-nano) and was replaced by none from gpt-5.1 on.
The page also covers two traps: the field must be nested as svg.reasoning.effort (a flat top-level field neither errors nor takes effect), and reasoning_tokens varies widely on the same simple prompt, so it cannot prove a tier took effect.
📖 Read the SVG reasoning FAQ
← Back to live updates · 📚 Monthly archive