OpenAI adds long-context support to GPT-5.6 Fast mode, doubling costs above 272K tokens

·1 views

OpenAI updated its API on August 5 to allow GPT-5.6 models — Sol, Terra, and Luna — to process prompts exceeding 272,000 tokens in Fast mode, which was previously called Priority processing. While Fast mode can deliver responses up to 2.5 times quicker, it carries an additional per-token premium on top of already elevated long-context rates. Prompts surpassing the 272K token threshold are priced at twice the standard input rate and 1.5 times the standard output rate, and enabling Fast mode compounds that cost further. For example, a 300K-token request using GPT-5.6 Terra costs $1.38 in Standard mode but doubles to $2.76 in Fast mode. Developers are advised to benchmark both modes on fixed test cases, use feature flags for gradual rollout, and only enable Fast mode when the latency gains justify the added expense.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

How Vue 3 and pdf-lib Enable Browser-Based PDF Splitting by File Size

A developer tutorial on DEV Community outlines how to build a browser-based PDF splitter that divides files by target size using Vue 3 and the pdf-lib library. Unlike splitting by page count, size-based splitting is complex because PDF file size varies widely depending on image resolution, font embedding, and content type. The proposed approach uses greedy page accumulation, incrementally saving the PDF after each page addition to estimate byte size rather than relying on simple arithmetic. When a page would push the current chunk over the size limit, the tool closes that file and begins a new one, ensuring no pages are lost even if a single page exceeds the target. The finished tool packages all output chunks into a ZIP file and includes user-selectable compression levels and a progress indicator for handling large documents.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Developer Scraps Six-Month AI Project to Build Modular 'Circuit' Architecture

After six months of failed attempts to make version 2 of an AI system work, a developer pivoted to a new modular architecture called v3 following an unexpected breakthrough with small, specialized circuits. The key insight was that narrow, purpose-built circuits outperformed a general-purpose monolithic model at specific tasks, using far fewer parameters and training in seconds. Version 3 replaced the single-pipeline design with a fleet of specialized circuits — including PropertyCircuit, RelationalCircuit, CausalCircuit, and ChainReasoner — each independently trainable and deployable. A set of ten guiding rules, documented in a file called TRUTH.md, was established to govern future architectural decisions, drawn from lessons learned across prior development failures. The central brain layer was retained but demoted from the system's core to one expert among many, consulted only when its capabilities matched the task.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

NETO offers dev teams encrypted P2P local chat with no cloud or accounts

NETO is a peer-to-peer messaging tool built for development teams who communicate over a shared local network or VPN. It requires no central server, no user registration, and ensures no data leaves the local environment. The app uses multicast DNS to automatically discover other peers on the network, eliminating the need for manual IP or port configuration. Encryption is handled via X25519 key exchange, with ephemeral keys negotiated between each pair of users. NETO positions itself as a privacy-focused alternative to cloud-based tools like Slack for sharing sensitive information such as tokens, passwords, or architectural details.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

California AB 723 Requires AI Image Disclosure Links, But Few Actually Comply

California's AB 723, in effect since January 1, 2026, requires real estate listings using AI-altered images to display both a visible disclosure label and a link or QR code pointing to the original, unaltered photo. While most platforms have implemented the on-image text label, the second requirement — a publicly accessible comparison URL — is widely overlooked. A developer building a virtual staging product found that compliance links were rarely clicked because they were buried in listing descriptions rather than embedded directly on the image. The law covers alterations such as added furniture, sky replacements, and landscaping changes, but exempts basic adjustments like exposure correction, cropping, and sharpening. Four other U.S. states are reportedly considering similar legislation, making proper implementation increasingly relevant for real estate tech products.

0 comments Read more at DEV Community