Models
66 RohitAI articles on Models: news, explainers and practical guides, newest first.
How SeeEEG gets image AI to classify brainwaves
For EEG researchers, a new preprint shows how signal images can reuse a frozen vision model—and why display design deserves its own test.
GitHub Copilot sets December 14 retirement date for six models
Six Copilot models are due to retire on December 14, 2026. Check the replacements, Haiku annual-plan exception and changing usage rates.
Google launches Gemini 4 Hydrogen with free API access
Gemini 4 Hydrogen offers free and paid access for business tools. Data-use terms, reasoning charges and shared context limits shape the choice.
Baidu opens ERNIE-Wizshot-ME for short talking-character videos
Baidu’s Wizshot makes short scripted character videos on Qianfan. Here are the required images and voice references, list-price costs and production limits.
Baidu opens ERNIE 5.5 Flash preview, with downloadable weights still to come
ERNIE 5.5 Flash Preview offers hosted access for coding and creative workflows. Here are the prices, tool requirements and promised weight-release date.
Choosing BotPsy for a simple local AI assistant
BotPsy offers a small model for local assistant prototypes. Compare its downloads, reported accuracy and limits before choosing it for an app.
Google opens Gemini 4 Carbon to paid users, with standard API rates set to double in December
Gemini 4 Carbon reaches paid developers and Ultra subscribers. Standard API rates double on December 1, and internal reasoning counts toward the bill.
Meta launches Muse Ember for routine AI work, with higher rates for long requests
Meta's new AI model targets recurring document, support and coding work. How to judge task fit and budget for longer requests.
Claude Fable 5.2 and Mythos 5.2 launch with 20% lower API rates
Anthropic cuts standard API rates for Fable 5.2 and Mythos 5.2. Here is who can use them, what they cost and what businesses should check before switching.
LG opens free K-EXAONE 2.5 access for non-commercial use
LG’s hosted K-EXAONE 2.5 is currently free, but non-commercial terms and request limits narrow how developers and researchers can use it.
Rafay adds model switching when AI budgets run out
Token Factory customers can stop requests, keep serving or switch to a cheaper AI model at a budget limit. The choice affects both costs and service.
What businesses need to self-host Kimi K3.1 Pro
Kimi K3.1 Pro’s public weights offer deployment control. Check hardware, commercial terms and workload costs before choosing self-hosting.