AI — Page 5

UST Chief Solutions Officer Kailash Attal tells RuntimeWire how Codon brings plain-English building to five major enterprise platforms while keeping every deployment behind approval gates, audit trails and mandatory human review.

Tuomas Artman's team cut runner time per test roughly in half while Linear's test suite nearly quadrupled in 2026, according to Linear.

The released GLM-4-Voice derivatives score up to 77.1% on GSM8K, though larger cascaded systems still lead the benchmark.

Parakeet Redux runs at a claimed 113x realtime on eight CPU cores, with sharper multilingual scores and a measurable penalty in noisy audio.

AJ Asver's internal test reduced compliance model costs from $2.89 to $0.25 per alert, with uncertain cases routed to a frontier agent.

The AI safety startup claims up to 2.8x TRL throughput, while its modified Apache license adds conditions for larger commercial users.

The Rome AI lab trained its first model in six weeks and released the weights, base model and vLLM serving code under Apache 2.0.

Code2Video tests agents on 168 motion-design briefs, then uses HeyGen's own judge model to score timing, composition and craft.

Morgan Linton's 92-run sweep found lower effort made Meta's model slower and less accurate, while Astra and Fable 5.1 fully passed all 23 tasks.

The dispute puts platform permission, customer credentials and Amazon's $68.6B advertising business at the center of agentic commerce.

The larger model targets longer coding and knowledge-work agents, with API rates of $2 input and $6 output per million tokens.

Ahad Ali built Tabby from his own CPA firm's workload; the startup reports 5,500 users and roughly $100,000 in ARR 14 months in.

The plugin generates, segments and displays 3D assets in chat, making OpenAI's generalist model the orchestrator for Meshy's specialist APIs.

The method searches interacting block-removal choices instead of ranking layers one at a time, with its biggest gains appearing under severe compression.

The local Python project stores every conversation line with a timestamp, turning one creator's years of assistant frustration into code.

Roger Jiang's model transfers human-action pretraining across LightBot and Unitree G1, while the open checkpoint is a 6B text-to-motion preview.

Jay Chooi's Robocurve tested three AI systems across 300 robot-arm trials involving five dangerous instructions.

Roy Horowitz's five-person startup joins a 23-company cohort while testing whether buyer-side car negotiation can work at one-fifth of human concierge fees.

The Cognition VP of Engineering released 0.8B, 4B and 9B models that return typed probabilities through an open, Jev-compatible local API.

Kling raised $2.8B after topping RMB850M in quarterly revenue; Sora's closure also erased Disney's planned $1B investment.

Dhravya Shah's black-box teardown says the personal agent relies on daily files and keyword search, then rebuilds the pattern in 60 lines.

The unified 2K generator accepts 10 reference images, while its license requires a separate agreement for any commercial use.

The service pairs Hugging Face web-seeds with peer discovery, but its 669,000-model catalog is far larger than its set of witnessed copies.

The January ALMA demo turned Andreas Hellander and Salman Toor's federated-learning research into onboard targeting, with a pilot held in reserve.