Claude Haiku 5.5: Cheaper AI for Many Small Tasks
Anthropic cuts API prices with Haiku 5.5. Where the small AI model helps, and why tokenization, prompt length, and thinking effort affect real costs.
Anthropic cuts API prices with Haiku 5.5. Where the small AI model helps, and why tokenization, prompt length, and thinking effort affect real costs.
OpenAI agents edited wikis, tampered with a citation tool and flooded Wikidata with requests. What Wikimedia found and what it means for the open web.
Mistral Large 4 launches as an API preview. What it offers for images and documents, and why downloading its model weights will have to wait.
Aleph Alpha has released Kolibri openly: 78 billion parameters, German as a core training language, Apache license. How it compares and who can run it.
Gemini lets mobile users attach a map area to questions. How Map works, what Google confirms, and which limitations still apply to location-based requests.
Audible is testing character guides, AI conversations, and images for audiobooks. What the features promise and why German users still have to wait.
Astra Ultrafast speeds up token generation. What developers should know about access, API pricing, EU processing restrictions, and actual workflow time savings.
Google is reorganizing the Gemini app: without a plan you get Flash-Lite only, AI Plus loses the Pro model, and AI Pro gains Deep Think. What changes, and for whom.
Microsoft adds streaming transcription and two voice models to MAI. What live captions, pricing, and Azure public preview status mean for developers.
Claude for Government is generally available. What the desktop app offers, which data reaches the cloud, and how agencies manage access and spending.