🤖 AI HOT Daily · Aug 4, 2026
Qwen3.8-Max launches with 2.4T parameters, first open Qwen-Max-level weights; SenseTime releases SenseNova U1.5-Lite lightweight multimodal model; OpenAI unveils GPT-Live real-time audio architecture that listens while speaking; Microsoft open-sources Orchard agent training framework; Cloudflare launches @cloudflare/computer agent runtime and Billable Usage API; EU AI Act transparency rules take effect with fines up to €15M; AirLLM runs 70B models on a single 4GB GPU; Gary Marcus says Anthropic mathematicians reproduced half of OpenAI Astra's results in 24 hours.
1. Qwen3.8-Max: Strongest Open Coding and Collaboration Model
Qwen's strongest model yet, with 2.4T parameters (95B active), the first Qwen-Max-level open weights, to be released next week.
2. SenseTime Releases SenseNova U1.5-Lite-Preview Open Model
A lightweight native unified multimodal model on NEO-Unify architecture, reaching commercial closed-source generation and editing quality with just 8B-MoT parameters.
3. GPT-Live: New Architecture for Real-Time Audio
GPT-Live can listen while speaking. Rebuilding the voice stack from client to model keeps audio flowing so deeper reasoning and tool use don't interrupt the conversation.
4. Microsoft Open-Sources Orchard Agent Training Framework
An open framework for the research community to train and evaluate AI agents across task types, reducing complexity so even smaller models perform strongly.
5. Cloudflare Launches @cloudflare/computer Agent Runtime
An open-source agent runtime providing each agent a virtual file system and supporting code execution in isolates, container sandboxes, or browsers.
6. EU AI Act Transparency Rules Take Effect
Effective Aug 2, companies must disclose when users interact with AI and machine-readably mark synthetic media, with fines up to €15M or 3% of global turnover.
7. AirLLM: Running 70B Models on a Single 4GB GPU
An open-source project runs 70B-parameter model inference on a single 4GB GPU without multi-GPU or large memory setups, gaining traction on Hacker News.
8. Gary Marcus: Anthropic Reproduces Half of Astra's Results in 24h
Anthropic mathematician Levent Alpöge reproduced half of OpenAI's claimed results in 24 hours using the public Fable model, raising doubts about Astra's breakthrough status.
9. Palantir CEO Calls Frontier AI Labs 'Marxist'
Alex Karp warns frontier labs seek to 'seize partners' means of production.' Palantir Q2 revenue hit $1.9B, up 93% YoY, championing model-agnostic AI and distributed approaches.
10. Cloudflare Launches Billable Usage API
A single endpoint returns usage and cost broken down by product and billing period for self-serve accounts, covering Workers, R2, D1, Workers AI, Vectorize, Images, and Stream.
Sources and verification
This legacy briefing did not retain its original source URLs. Verify safety, financial, policy, and breaking-news claims with authoritative primary sources before acting on them.