Artificial intelligence, explained in plain language.
Explore Maitz articles about API: practical guides, news and clear explanations.
According to DeepSeek, V4.1-Flash-1017 improves coding and agent scores, shortens outputs, and lowers cost per task, with API prices unchanged. Weights are MIT-licensed, and Search and Fetch are now open to API users.
According to Anthropic, Claude Opus 5.5 fast mode delivers up to 2.5x output tokens per second at premium pricing, and no longer needs an access request.
According to OpenAI, GPT-Live-1.1 adds 13 voices and better French, German, Hindi, Korean and Japanese. The API is listed as coming soon.
According to NVIDIA, Super VL is a 120-billion-parameter open model for agents, and Bolt 74B runs locally on NVIDIA hardware. Both are available now; Bolt E4B ships October 23.
According to Microsoft AI, MAI-Base-2 beats every open-weight base model it tested. Post-training is still underway, with weight licensing planned for trusted Foundry partners.
SpaceXAI’s new voice model tops the Artificial Analysis speech-to-speech index, replies in under a second and keeps the same price as its predecessor.
Memory that carries across conversations, a 1M-token context and a model built to push through hard tasks: Meta aims its new generation at agents that do real work.
Anthropic has released Claude Haiku 5.5, a small, fast model for high-volume work such as summaries, classification and subagents. The company says it costs about 75% less to run than Haiku 4.5 on average. Sonnet 5.5 cache reads also got cheaper, and Max and Team subscribers get monthly API credits.
OpenAI has released GPT-6.1 Sol: a model that approaches Astra on coding, PDFs, and computer use at about one-fifth the cost. Here is what the official announcement says about pricing, availability, and evaluations.
