🤖 AI HOT Daily · Aug 14, 2026
Xiaohongshu open-sources dots.tts, a 2B-parameter fully-continuous autoregressive TTS model; Google DeepMind ships Gemini 3.7 Flash for coding and agents at half price; MiniMax releases Music 3.0 with up to five-minute songs; DeepSeek Harness v0.1 dev preview goes open source; Cursor launches builds (3x faster cloud-agent startup) and earns AIUC-1 certification; OpenAI previews Ultrafast, running GPT-5.6 Sol up to 14x faster; OpenAI names Dali Rajic Chief Revenue Officer.
1. Xiaohongshu Open-Sources TTS Model dots.tts
A 2B-param fully-continuous end-to-end autoregressive TTS model topping Seed-TTS-Eval on average content accuracy and speaker similarity.
2. Google DeepMind Launches Gemini 3.7 Flash
Just three weeks after 3.6 Flash, it targets coding and agent tasks at $0.75/$3.75 per million input/output tokens — half the previous price.
3. MiniMax Music 3.0 Released
A new open-weights, production-grade music model composing, arranging, performing, and producing full songs from a concept plus optional lyrics, up to five minutes.
4. DeepSeek Harness v0.1 Dev Preview Released
An agent framework on the Cordis meta-framework, MIT-licensed, with an everything-is-a-plugin core — models, tools, skills, sandboxes, orchestration, and UI all composable.
5. Cursor Builds: Cloud Agents Start 3x Faster
Always-ready environment copies in the background let cloud agents skip cold builds, cutting startup latency up to 3x; default on for all environments from Aug 17.
6. OpenAI Previews Ultrafast Mode
A new Cerebras-powered API tier runs GPT-5.6 Sol up to 14x faster at up to 750 output tokens/second — currently in preview.
7. Claude Code v2.1.232 Enables Subagent Forking
Subagents inherit full context and prompt cache, non-teammate agents run in background by default, with new GitLab support and Windows security fixes.
8. Cursor Earns AIUC-1 Certification
Cursor passed independent review and adversarial tests under a standard shaped by 100+ Fortune-500 CISOs, covering IDE and cloud agent safeguards.
9. GPT-5.6 Builders Guide: Frontier Agent Performance for Less
New API capabilities include reasoning persistence, native multi-agent orchestration, and programmatic tool calls; on ARC-AGI-3 Sol jumped from 13.3% to 38.3% with ~6x fewer output tokens.
10. Claude Handles App Maintenance: 388 PRs
Boris Cherny let Claude run crash fuzzing, dedupe, and dead-code removal via Slack, auto-opening 388 PRs in weeks, with 180 merged after review.
11. Anthropic Studies Emerging Multi-Agent Systems
45 coordinated agents found 266 vulnerabilities in 27M tokens, vs 21 for independent parallel methods in 6.5M tokens — only 12 overlapped.
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.