ZD Tech
AI Daily

🤖 AI HOT Daily · Jun 29, 2026

Grok 4.5 tested at SpaceX and Tesla approaches Opus performance, Sina open-sources VibeThinker-3B showing reasoning compresses but knowledge doesn't, Wayfinder Router enables deterministic LLM query routing, Adrafinil prevents Mac sleep only during AI agent work, only 3 AI models profitable in 500-day startup simulation, top AIs play Civilization VI exposing perception gaps, open-source model ecosystem expands with NVIDIA and Cohere releases.

  1. 1. Grok 4.5 Tested at SpaceX & Tesla, Nears Opus Performance

    Based on 1.5T V9 base model with Cursor data in supplemental training. Early evaluations show performance near or surpassing Opus. SpaceX to release monthly retrained models.

  2. 2. Sina Open-Sources VibeThinker-3B: Reasoning Compresses, Knowledge Doesn't

    Only 3B params, matches DeepSeek V3.2 (200-333x larger) on AIME26 benchmarks, but falls behind on knowledge-heavy GPQA-Diamond. Proposes 'parameter compression-coverage hypothesis.'

  3. 3. Wayfinder Router: Deterministic LLM Query Routing

    Microsecond-level routing decisions by analyzing prompt structure. Fully offline, no external model calls. Supports any OpenAI-compatible API, self-hostable.

  4. 4. Adrafinil: Prevent Mac Sleep Only During AI Agent Work

    Supports 9 AI coding agents including Claude Code, Codex, Cursor, Gemini CLI, Aider. Mac sleeps normally when no agent is running.

  5. 5. Only 3 AI Models Profitable in 500-Day Startup Test

    Princeton's CEO-Bench: of 14 models, only Claude Fable 5 ($47.15M), Opus 4.8 ($27.8M), and GPT-5.5 ($21.3M) exceeded $1M starting capital. Most went bankrupt before the simulation ended.

  6. 6. Top AIs Play Civilization VI: Exposing Perception & Execution Gaps

    Claude, GPT-5.4, Gemini 3.1 Pro play 23 games of Civilization VI. AI proactively checks game state only 1-2% of turns. Plan execution rate within 10 turns: just 48-66%.

  7. 7. Artifacts 22: Open-Source Ecosystem Expands with NVIDIA & Cohere

    Open-source model landscape diversifies globally. NVIDIA releases Nemotron-3-Ultra-550B, Cohere open-sources flagship model under Apache 2.0.

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.