AI — Page 19

Kushal Byatnal's new API flags synthetic documents and post-signature edits, though Verify's reported 96.3% recall comes from a private internal benchmark.

Stefano Ermon's diffusion model claims a 1-million-token context window and benchmark gains over rival models, with launch pricing half off.

The October 18th release adds a $4-per-1,000-request tier and token-saving highlights; Exa's benchmark leadership claim comes from tests it ran itself.

The order, signed July 24th and posted for Federal Register publication on October 20th, also covers third-party products built with DeepSeek implementations.

Founder Tarini Sai Padmanabhuni says the API and MCP service checks audio in 24 ms; a benchmark found 94.47% accuracy and an 8.4% false-positive rate.

Arena.ai says the model scored 1,726 points on October 19th, a week after Meta released it for subscribers of Muse Code's higher-priced plans.

Adi Singh and Sven Myhre pitch Reason as a team product, with cloud agents, model flexibility and shared worktrees at its center.

Modal is testing an Arm server for its burst-oriented sandbox fleet, where CPU contention can slow agents and leave training GPUs idle.

The reported 18 configuration changes have not been independently verified; public OpenAI and Microsoft documents list GPT-6.1 Sol, not Terra.

DeepSeek released V4.1 Flash 1017 on October 17th; Together AI said it was live through its API on October 19th. The listing prices it at $0.30 per million input tokens and $1.20 per million output tokens.

Treg says the toolkit records sites cited by AI systems and identifies their authors; its broader product also gives agents access to third-party tools.

Co-founder Mohammad Abu Sheikh is leading the UAE-based startup's expansion. CNTXT AI says it has $300 million in "committed revenues," mostly from U.S. buyers, and plans 15 to 25 commercial hires in its first year.

Agencies can start with hands-on workshops at Accenture's Washington, D.C., Forge and Prototyping Lab, then work with embedded engineers on Gemini agents. The announcement names no pilot agency and specifies no launchpad-level security authorization, pricing, or procurement route.

Higgsfield says its open-source CLI can call more than 35 models, automate batch jobs and charge generations to existing plan credits.

Amit Rath's open-source project puts a large model in the setup step and a small local model on the job, while its Pac-Man demo shows the limits of that split.

The team reports lower token use and faster inference across 11 benchmarks, while its cost savings remain estimates that exclude prompt processing and tool calls.

A paper led by Zhanli Li and researchers at Meituan and universities uses explicit causal rules to generate messy files and score data agents against the hidden record.

The proposal needs stockholder approval. Anthropic says the fund could receive more equity as Anthropic grows and estimates roughly 20% of its shares will support public benefit.

The study, published on October 18th, put six personal-assistant agents in a simulated refund dispute and tracked what they did after the claim deadline passed.

The draft sets a threshold of more than 50 people killed or seriously injured, or more than $1 billion in damages, before Commerce can ask a court to halt a release.

DeepSeek opened Search and Fetch APIs on October 17th, making its in-house search engine available to agents and retrieval apps at $0.75 per 1,000 successful requests.

Lisan al Gaib shared the identifiers on October 18th, but provided no API response or evidence that developers can call them.

The November 12th-13th San Francisco program lists leaders from Intel, OpenAI, Thinking Machines Lab, Perplexity, Google, NVIDIA and AWS, as SGLang's serving framework continues to ship updates.

The October 16th update adds shareable canvases, a beta Environments API and support for larger, faster-loading cloud-agent environments.