TLDR AI 2026-07-30

Thinking Machines co-founder Lilian Weng left the startup after citing health effects from sustained stress and workload, then joined OpenAI. She said the pace required by the startup had become physically unsustainable.

Grok Voice Think Fast 2.0 is now available at $0.09 per audio minute. grok-voice-latest will switch over to the new model on August 5. The release makes Grok Voice more dependable in real customer workflows.

GPT-5.6 Sol scores just 7.8% on the ARC-AGI-3 benchmark despite having solved longstanding open problems in mathematics and beaten games like Pokémon FireRed. Benchmarks rarely measure AI models in isolation. They also measure less visible choices about API settings, harness design, and prompting. Researchers discovered that turning on retained reasoning and compaction in ChatGPT and Codex tripled scores and cut output tokens by 6x on the benchmark.

Google's latest music generation model, Lyria 3.5, is now rolling out in Google Flow Music. The model delivers significant advancements across musicality, lyrics, and vocal quality. A clip of a song generated by the model is available in the article.

Escha-W2 is a 2-bit quantized build of Qwen3.6-35B-A3B, a Mixture-of-Experts model with 256 experts. The model is packaged with everything needed to serve it locally through an OpenAI-compatible HTTP API. The whole thing is 12.3 GB on disk and runs on a single 24 GB consumer GPU - or on a 16 GB card.

Liquid AI released encoders designed for efficient document-scale inference on CPUs. This article covers their 8,192-token context window, competitive benchmark results, and their lower long-context latency.

DeepMind researchers suggest that visual Prompt Engineering can improve video-model reasoning by transforming task images before inference, such as converting abstract sketches into photorealistic scenes.

Parallel Decoding Distillation is a trajectory-based method that predicted multiple denoising steps during each model evaluation. It achieved state-of-the-art results with four to eight evaluations across several image and video generators while improving video diversity.

500,000 people have already signed up for TLDR Hardware, our new twice-weekly newsletter covering chips, robotics, energy, and devices. If you work in hardware and want to help curate it, send your LinkedIn or resume to hardware@tldr.tech!

Claude Opus 5 is the top-scoring model on Vending-Bench, a vending machine simulator. The model discovered that focusing on higher-end products yielded higher profits, and it never gave a single dollar to scammers. It demonstrated some misaligned behavior, such as fabricating competitor quotes when negotiating with suppliers and lying about delivery delays. The model also proposed or engaged in price cartels in all runs - most cartels ended with Opus breaking the truce and undercutting the others.

Compute costs may rise 10x as AI labs like Anthropic aim for $1 trillion revenue, driven by increasing margins, rising compute prices, and more spending on inference. Google pays twice the spot price for GPUs due to demand, with stronger monetization of AI models leading to higher compute value. High compute costs may prioritize efficient AI, pricing out less critical applications and intensifying competition in AI development.

Cargo, Grepsr, and Dust each hit their own wall scaling AI reliably. This free Temporal eBook covers the real architecture behind how they fixed it. Get your copy

Requesty is the gateway for production AI: 600+ models, failover, auto-caching, analytics & governance behind one API.

Route your first call (free)

Google is developing an artifact type for Gemini Notebook to transform sources into interactive apps, introducing a new "App" tile feature.

A harness should capture what the human actually wants, convey it to the model on every task, and otherwise stay out of the way.

Deep Agents v0.7 reduces base input tokens by 65% while maintaining performance, improving token and cost efficiency.

Pangram is an AI detector that achieves roughly one false positive for every 24,000 documents.

Numbat, an open-source security suite by Perplexity, addresses security risks in AI agents deployed on client endpoints by integrating with agent harnesses to prevent, detect, and mitigate incidents like accidental meltdowns.

Moonshot is now reaching out to potential backers for a new funding round at a $50 billion pre-money valuation.