Tech • AI • Robotics • Game
OpenAI Decisions API debuts as Base44 tops Replit, Bolt, Lovable
Briefing
OpenAI unveils Decisions API
OpenAI introduced a new Decisions API aimed at software workflows that need probabilities, labels and routing signals rather than generated prose. The product evaluates both text and images, positioning it for business automation such as spam checks, urgency triage and ticket classification. Instead of returning paragraphs, it emits structured decision outputs that applications can act on directly. The launch marks a clear product split between generative interfaces and low-latency decisioning systems.
GPT-6 Luna powers classification
The new API currently runs on GPT-6 Luna and exposes three decision formats: predicate, score and choice. Predicate estimates whether a statement is true, score rates an input across ordered levels, and choice selects from predefined options. That design targets common enterprise patterns such as routing, prioritization and moderation. By constraining outputs to structured decisions, OpenAI is pushing model behavior closer to conventional software logic.
OpenAI targets speed and price
OpenAI says the Decisions API is roughly 10 times faster than the Responses API for simple evaluation tasks. Pricing is listed at $0.10 per million input tokens, with no output-token charge because the system does not generate text responses. That could materially lower costs for high-volume classification pipelines. The economics make the product especially attractive for short, repetitive backend workflows.
Base44 wins CRM builder test
In a head-to-head build test, Base44 ranked first ahead of Replit, Bolt and Lovable after generating the same CRM under identical rules. The benchmark required three stages: an initial dashboard, a notes and follow-up upgrade, then search, mobile adaptation at 375 pixels and empty-state fixes. Platforms were scored out of 100 across speed, learning curve, quality, flexibility and value. The result adds to the growing scrutiny of AI app generators on stability, not just first-pass demos.
Base44 posts 96 out of 100
Base44 finished with 96/100, including 20/20 for speed, quality and flexibility. It reportedly produced a first working CRM in about 4 minutes, then completed two additional iterations in 4 minutes and 3 minutes without major failure. The generated app also included extras such as dark mode, a recent-clients panel, an activity feed and sample data. That combination of speed and robustness set it apart from the rest of the field.
Claude Pro math shifts
Guidance around the $200 Claude subscription is coalescing around weekly token management rather than aggressive context trimming. Reported limits make short session caps manageable, but weekly all-model ceilings can halt work for days if users burn through quota early. A practical rule of thumb is roughly 14% per day across a seven-day cycle. The advice reflects how premium model subscriptions increasingly behave like managed compute budgets.
Fable pricing pressures Claude users
Among the models discussed, Fable is framed as the most expensive option for routine work at about $10 per million tokens and roughly 2.5 times the cost of Opus. Sonnet is positioned as the cheaper everyday choice, with pricing cited near $2 input and $10 output per million tokens. The key warning is that output and hidden reasoning tokens, not just prompt length, drive consumption fastest. For heavy users, model selection and reasoning mode discipline matter more than shaving a few lines from context.
Automation pushes short-form publishing
A separate workflow shows how AI tooling is moving beyond coding into automated media operations, turning long recordings into short clips for Instagram, TikTok and YouTube Shorts. The system reportedly handles editing, subtitles, motion design, covers, posting and even DM replies, while reusable style files preserve brand-specific colors, fonts and framing. One newly launched Instagram account posted 32 automated pieces and saw individual reels reach roughly 1,400, 1,600, 4,400, 3,100 and 1,800 views. The broader signal is that AI-native production stacks are compressing both creative and distribution labor into a single pipeline.
Videos covered
Decisions API Full Tutorial for Beginners: How to Use & Set up Decisions API
- •A shift from generation to decisioning
- •Faster and cheaper for simple workflows
- •One model, three decision types
Comment AVOIR Claude en illimité et arrêter de cramer tes limites
- •Weekly limits matter more than session caps
- •Tokens drive the real cost
- •Fable is the costliest option
Je remplace TOUT par Claude (ManyChat, Social Media Manager) et cet outil cheaté
- •Early results on new accounts
- •AI handles nearly all editing
- •Style files replace one-size-fits-all templates
Best AI App Builder in 2026 (Ranked Worst to Best!)
- •A methodology focused on stability
- •Base44 dominates through speed and robustness
- •An almost flawless final score for Base44