AI — Page 30

Cloudflare's open-source preview gives agents a durable filesystem while reserving full Linux environments for heavier tasks.

Yannik Marek's node-based engine added native MiniMax H3 workflows, giving creators inspectable graphs for the open-weight video model.

Invenio Imaging enrolled 1,006 patients in its U.S. pivotal study of NIO Lung Cancer Reveal, collecting more than 3,250 fresh biopsy specimens for validation of the investigational AI system.

Founders are building measurement, routing and infrastructure products for buyers that adopted costly frontier models before tracking their returns.

Artificial Analysis put Liang Wenfeng's model near Gemini 3.6 Flash on intelligence and roughly 105 times cheaper than Claude Fable 5 in benchmark test cost.

The 2.4 trillion-parameter model is live on QwenCloud at $2 per million input tokens, while downloadable weights are still pending.

The 288 GB Hugging Face release adds two downloadable base checkpoints, while the complete 2K workflow still uses hosted MiniMax services.

The hosted preview pairs a 1 million-token context window, built-in tools and subscription credits as Alibaba pushes Qwen into developer workflows.

QM gives each employee and shared room a durable computer, memory, credentials and background jobs, then routes four different coding-agent harnesses through one policy core. The open release is ambitious, legible and unusually candid about where its security model breaks down.

Cloudflare's April Agents Week bundled its serverless, security and network products into a bid for autonomous software workloads.

MLX.fast turns each verified inference improvement into the next baseline, extending Eigen's open research system beyond quantum cryptography.

Wafer reports 952 tokens per second from one eight-GPU MI355X node, arguing AMD's 288 GB-per-GPU memory can lower open-model inference costs.

MiniMax launched H3 with 2K video and native stereo audio. MiniMax says the weights are coming in days, but the H3 license and official local hardware requirements remained unpublished as of August 2nd.

The iRobot co-founder says research, hype, deployment and economic change move at different speeds, challenging near-term automation bets.

The site packages model-price comparisons into calculators for APIs, chatbots, agents, retrieval systems and other AI workloads.

Independent benchmarks and access to 3.60 billion daily users undercut claims that Meta has little to show for AI, but a capital budget of up to $145 billion leaves the business case unproven.

The model doubles single-pass length, adds dense reference inputs and rolls out through ByteDance's own AI products, with an API planned.

Dylan Ayrey's couch-born TruffleHog found reused keys across public AI corpora, where deleting one source cannot erase its copies.

The proposed class action says non-users cannot consent or opt out when Granola transcribes calls and uses anonymized data for model improvement.

The lab released a 249-page paper, reasoning transcripts and Lean certificates for claims spanning geometry, cryptography and quantum complexity.

DeepSeek's open-weight model undercuts rivals on API price, while third-party tests put its max-reasoning version near the top of its class.

Lon says a Claude Code path can route requests to Opus 4.8 without the notice and opt-out Anthropic's own help page promises.

Sam Altman demonstrated the unreleased model to policymakers as a federal review process approaches its August 1st deadline.

The open-source tool lets agents map, search and read slices of Markdown knowledge before modifying a codebase.