OpenAI Cuts GPT-5.6 Prices and Adds Fast Mode to Expand API Options

·59 views

OpenAI has announced pricing and performance updates to its GPT-5.6 model lineup, reducing Luna's cost by 80%, Terra's by 20%, and introducing a Fast mode for Sol. The company frames these changes as an effort to advance what it calls a price-performance frontier, giving developers more tailored options across different workloads. CEO Sam Altman has publicly stated the goal of offering the best intelligence-to-cost ratio at multiple points along a trade-off curve, rather than relying on a single flagship model. The updates signal a shift toward more operational model selection, where cost, speed, and output quality are weighed together depending on the use case. OpenAI has not specified a universal formula for choosing between its GPT-5.6 variants, leaving teams to evaluate the options against their own production requirements.

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)

Pharo Smalltalk Runs in the Browser via WebAssembly

A project has been shared on Hacker News showing the Pharo programming environment running directly inside a web browser. The implementation uses WebAssembly to execute Pharo's virtual machine client-side. The demo is hosted as a web page on a personal GitHub Pages site. The post drew modest attention, with 4 points and no comments at the time of listing. The source material provides limited detail beyond the link itself.

0 comments Read more at Hacker News

ProgrammingDEV Community ·

Outlook Search Missing Emails? Check the Windows Search Index First

Help desk staff often find that Outlook search fails to show emails the user is sure exist, and the usual first step of rebuilding the Outlook profile may be unnecessary. In classic Outlook for Windows, search depends on the Windows Search index, so a paused, stalled, or corrupted index can hide messages that are still in the mailbox. Technicians can check indexing progress under File > Options > Search > Indexing Options, and confirm the WSearch service is running via PowerShell. If the problem persists, rebuilding the index from the Advanced settings is recommended, though it can take hours on large mailboxes. Finally, verifying the cached mail download range and comparing results with Outlook on the web can reveal whether the issue is local.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

AML Analysts Lose Capacity When Case Retrieval Is Manual

A DEV Community post frames slow manual retrieval of case information as a major defect in anti-money laundering operations. It argues that analysts lose substantial capacity when they must gather history, links, and background themselves before reviewing an alert. The author describes a common workaround of asking a senior colleague, which fails once that person leaves the organisation. The proposed fix is to retrieve relevant sources before an analyst opens a case, present them transparently, and keep human decision-making in charge. The piece characterises this as a workflow problem rather than a model-accuracy issue.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Acceptance testing a small salon app with fake data across 61 scenarios

Before a real salon used the app, a separate AI session acted as the owner on a test computer with fake data, running 61 scenarios over four rounds. The first round uncovered 19 defects, some serious, which were all fixed, and later rounds came back clean. The scenarios targeted difficult cases such as forced password changes, lockouts after wrong passwords, messy Italian-language spreadsheets, duplicate imports, wrong time zones, data isolation between salons, large customer lists, unreliable networks, and mobile and keyboard use. The report also noted what could not be verified, such as the real booking-software export and WhatsApp messages, and left these for a supervised first session with the owner. The author now runs such an acceptance round before handing over any app, since it catches a different kind of mistake than automated tests do.

0 comments Read more at DEV Community