Realtime AI News
TokenRhythm opens limited beta for its 180B agent-focused NeoHorse-1.5-Flash
TokenRhythm has opened a limited beta for NeoHorse-1.5-Flash Preview, a 180B-parameter model positioned for agent workloads, according to Pandaily. The model is being offered in preview form, with details on access scope and availability still thin.
TokenRhythm has opened a limited beta for NeoHorse-1.5-Flash Preview, according to a report from Pandaily. The model is a 180B-parameter system positioned for agent workloads and is being offered in preview form.
Three words carry most of the information here: limited, preview and 180B. A limited beta usually means access is granted in batches or on request, while preview status signals that the model is not final and that its interfaces, behavior and performance may still change.
The 180B size sits in the large-model bracket: far beyond what runs on a single consumer GPU, yet not the extreme scale of the largest dense models. That range is commonly used for general or specialized workloads that need solid reasoning and tool-calling ability.
The agent framing is the most concrete claim in the announcement. Agentic products depend on a model's ability to call tools reliably across multiple turns, follow structured instructions and hold task state, so naming and positioning a model around agents effectively declares which workloads it is being tuned and evaluated for.
TokenRhythm is a relatively new model provider, and public information about NeoHorse-1.5-Flash currently comes mainly from Pandaily's report. Beta periods are also when vendors gather real-world feedback and fill gaps in documentation and tooling, so availability, rate limits and integration paths may matter more right now than the parameter count.
Three things are worth watching: whether beta capacity expands, whether public benchmark results appear, and how open the weights and API access turn out to be. Those answers will decide whether NeoHorse-1.5-Flash becomes something developers can actually build on or remains a demonstration.
Why it matters
A 180B agent-specific model entering limited beta shows vendors tuning for agentic workloads rather than generic chat, though preview-stage access and transparency still need watching.
Nearby Updates
All11/09, 15:00
DeepSeek's Harness Agent Tops 1.5 Million Daily Users
DeepSeek's AI agent push is gaining pace, with its Harness product now exceeding 1.5 million daily active users, according to the South China Morning Post. The figure shifts the conversation from model benchmarks to real usage at the consumer-facing end of the agent race.
11/09, 14:00
Wirken: an open-source gateway that assumes every AI agent can be compromised
Wirken is an open-source gateway that sits between messaging apps and AI agents and assumes any agent can be turned against its owner. It isolates each channel in its own process, keeps credentials in an encrypted vault, runs shell commands in a networkless sandbox and records every action in a signed, hash-chained log.
11/09, 13:06
NVIDIA Dynamo Recipes target production use for Korea's leading open models
A short NVIDIA blog item says Dynamo Recipes can bring South Korea's leading open models into production. The point is the deployment path rather than a new model, and the source gives no model list or performance figures.
11/09, 13:02
NVIDIA, Rebellions and SK Telecom announce collaboration
NVIDIA, Korean chip designer Rebellions and telecom operator SK Telecom have announced a collaboration. The available summary stops at the headline, with no stated scope, technical detail, investment size or timeline.