Exchange Ranking
Top Exchanges
24h Volume Ranking
Popularity Ranking
Exchange BTC Balance
Proof of Reserves
Decentralized Exchanges
Funding Rate
Funding Heatmap
Liquidation Data
Max Pain
Long/Short Ratio
Whale L/S Ratio
Binance/Okex/Huobi L/S
Bitfinex Margin L/S
ETF Tracker
Solana ETF
XRP ETF
Hong Kong ETF
Bitcoin Treasuries
Crypto Reversal
Ethereum Reserves
HyperLiquid Wallet Analysis
Hyperliquid Whale Watch
Large Transactions
On-chain Movement
Bitcoin ROI
Stablecoin Market Cap
Options Analysis
News
Articles
Economic Calendar
Features
Wallet
Contract Calculator
Security
Collections
Watchlist
FollowingOpenAI Prompt Guide Update: Streamlining system prompts can improve scores and significantly reduce Token consumption
OpenAI has now told developers that removing duplicate rules from system prompts can increase internal coding evaluation scores by up to 15%, while reducing Token usage by up to 66%.
Key Points
The GPT-5.6 Prompt Guide requires developers to define results and termination conditions, and then let the model choose its own path. Operation of the internal coding agent shows that streamlined system prompts increase evaluation scores by 10% to 15%, reduce Token usage by 41% to 66%, and reduce costs by 33% to 67%. This document adds "Programmatic Tool Calls" and text. There are two parts of detail setting, none of which appear in the old version of the GPT-5 guide.
OpenAI Prompt Guide for Rewriting GPT-5
OpenAI released the guide with the GPT-5.6 model series, which became fully available on July 9 and is mainly aimed at API developers and teams running automation agents. The document requires engineers to clarify what results, constraints, available evidence, and completion criteria are visible to the user, and then leave it to the model space to choose its own efficient path. This method is called "result-first prompting."
This recommendation is far from the GPT-5 guide released in August 2025, which promotes XML persistence blocks, detailed context collection templates, and tool prefix scripts that step by step describe each step. Today, these restrictions are regarded as noise. OpenAI also warns against using absolute rules such as "always" and "never", leaving them only in truly constant scenarios, such as security restrictions, required fields, and actions that must never be allowed to occur. The guidelines state that repeating instructions such as "inquire first" or "wait for approval" may trigger unnecessary approval requests for safe and expected operations. Conflicting rules are more likely to cause instability than missing details.
Independent developer Simon Willison pointed out that "programmatic tool calls" and multi-agent support are the most interesting additions to this release, a feature that allows the model to combine and run JavaScript to orchestrate tool calls. He also said Sol was competent enough in the complex coding tasks he had been performing, but did not significantly surpass Anthropic's Claude Fable 5.
Cost is the second reason why this guide is important. In internal operations, Token usage dropped by 41% to 66%, and expenses dropped by 33% to 67%, figures that change the costing of any team running agents on a large scale. OpenAI warns that results vary depending on workload and says these ranges serve only as directional references, requiring developers to verify the data on typical tasks for their own applications.
GPT-5.6 series fast iteration
GPT-5.6 is released in three specifications: Luna, Terra and Sol. The input prices are US$1, US$2.5 and US$5 per million Tokens respectively, and the output prices are US$6, US$15 and US$30 respectively. New text has been added to the guide. The detail setting is used to control the response length.
This shift follows a continuous pattern. OpenAI's April GPT-5.5 guide already requires teams to rebuild hints rather than porting old hints forward; while the August 2025 GPT-5 guide goes in the other direction, requiring clear limits to be set around "proactive level." Each cycle is pushing developers to write less and allowing models to take on more path choices.
