Latest

Every published item, in order · 9305 items
Oct 17
ModelsUS vs ChinaSelected08:11
Muse Spark Open: Meta Releases 816B-Parameter Open-Weights Model

An 816B-parameter MoE that activates only 42B per token and runs on two Mac Studios signals Meta pushing frontier-capable open weights into consumer-grade hardware territory. Builders comparing local deployment options should watch the license terms, which still limit some commercial use. The release gives OpenAI and DeepSeek's open-weight line a direct competitor at the 1M-token context level.

SuperTechFansSource
Tomorrow 08:00, the daily lands in your inboxTelegram channel or email. Same screening as this page, no ads.
ModelsBuildersMulti-sourceSelected05:16
GPT-6 Astra Pro upgrade speeds up ChatGPT and adds parallel agents

Parallel agents with per-agent progress tracking turn ChatGPT into a lightweight orchestration workspace, reducing the need to chain separate tool calls manually. Pro users in Singapore and across ASEAN can try it immediately in the Chat tab without any setup. Teams running research or multi-step workflows should compare it against single-thread agent tools like Cursor or Claude before choosing one.

Event
Official accountOpenAI@OpenAISource
ModelsBuildersSelected03:18
TypeSafe's Jev-1.14 Structured Decision Model Lands on Databricks

The 400x cost gap signals a shift: high-volume classification and scoring workloads may no longer justify general-purpose LLM calls, changing how teams architect real-time decisioning. Data engineers and ML platform teams running large Lakehouse pipelines can test Jev immediately via ai_classify and ai_decide without new infrastructure. Compared with standard LLM inference, its value is latency-bound batch scoring rather than open-ended generation.

Event
Official accountDatabricks@databricksSource
ModelsUS vs ChinaSelected03:05
Reflection AI's Beam 1.0 Enters Agent Arena Top 15 Labs

A US lab landing a top open-weight spot in agentic benchmarks breaks the pattern of Chinese labs dominating the open WebDev leaderboard, where 24 of the top 25 models are Chinese. Teams choosing an Apache 2.0 model for agent or coding work now have a credible US alternative to Kimi K2.5 Thinking and GLM-5. The wide error bars mean its true standing relative to Qwen3.8 is less settled than the rank suggests.

lmarena.ai@lmarena_aiSource
ModelsUS vs ChinaMulti-sourceSelected71°02:44
GLM-5.3 tops Epoch AI Cyber ECI among open-weight models

The 6–7 month gap is a quantified benchmark of how quickly open-weight models absorb frontier-level cyber offensive capability, a metric security teams and policymakers can use to set review timelines. Chinese open-weight models closing in on US closed models is the notable trend here, relevant to anyone comparing model sourcing for security-sensitive workloads across ASEAN and beyond.

Event
Official accountEpoch AI@EpochAIResearchSource
ModelsBuildersMulti-sourceSelected00:48
NVIDIA open-sources ASPIRE robot learning framework

Saturated simulation scores mean LIBERO-Pro and Robosuite no longer separate strong methods from weak ones, so real-robot results and continual refinement become the meaningful signals here. Robotics researchers and embodied-AI engineers can clone the code and benchmark their own stacking or manipulation pipelines against it. The GPT-6 Astra integration hints at a general model driving physical manipulation rather than task-specific policies.

Event
Official accountJim Fan@jimfanSource

Only the latest 2000 items are listed; for older ones see the daily archive