Crypto
Home›Crypto›Market Structure›OpenAI updates GPT-5.6 prompting guide to cut system p…
OpenAI updates GPT-5.6 prompting guide to cut system prompt bloat
OpenAI says internal coding-agent tests found leaner system prompts boosted evaluation scores by about 10 to 15 percent while reducing tokens by 41 to 66 percent and lowering costs by 33 to 67 percent.
OpenAI has published a dedicated prompting guide for its GPT-5.6 Sol model that encourages a shift away from lengthy, heavily scripted system prompts toward outcome-first instructions, according to Decrypt.
The guidance emphasizes defining the desired destination, setting stopping conditions, and avoiding unnecessary detail. It also points to a new section on Programmatic Tool Calling and highlights OpenAI’s text.verbosity API parameter, both described as absent from the earlier GPT-5 prompting approach.
OpenAI cites internal coding-agent testing results showing that lean system prompts improved evaluation scores by roughly 10 to 15 percent while cutting total tokens by 41 to 66 percent and costs by 33 to 67 percent. The guide frames the change as reducing noise such as repeated rules, style instructions that do not affect behavior, examples that do not change outputs, and process steps the model already handles reliably.
Decrypt reports that the earlier GPT-5 guide focused on adding scaffolding, including XML persistence blocks, context gathering templates, and tool preamble scripts to manage when to escalate or stand down. The new guidance warns that GPT-5.6 follows prompt contracts more closely, and that conflicting rules can create more instability than missing detail.