AI Tech News Today — with Aurora & Isabelle

Salute: listeners
Date: 2026-07-10T12:12:58.000Z

Today we’re rounding up the biggest AI platform moves, model clashes, and a few product tweaks that matter to developers and businesses.

Top headlines

OpenAI: GPT-5.6 family and Ultra Mode

• Model family: GPT-5.6 released in three tiers — Sol, Terra, and Luna — tuned to cut token costs and speed up agentic coding.
• Ultra Mode: Sol leads with an Ultra Mode that coordinates four parallel agents, reducing token use on cyber benchmarks by roughly one third.
• New products: ChatGPT Work (multi-step documentation across mobile and desktop) and GPT‑Live voice system, which supports natural interruptions by splitting voice and reasoning pipelines.
• Pricing: $1–$5 per million tokens, with Sol confirmed as the preferred model for Microsoft 365 Copilot.

Anthropic: lower context costs and Reflect

• Cost changes: Anthropic reduced usage caps and context costs for Fable 5.
• Reflect dashboard: Launched a beta Reflect (a screen-time-style report) to show users chat habits when Memory is on.
• Multi-agent routing: Introduced a framework that routes tasks to cheaper sub-agents to preserve coding power while trimming costs.
• Governance: Appointed former Federal Reserve Chair Ben Bernanke to an independent trust to study long-term economic impacts — signaling the changes are strategic, not merely competitive.

SpaceXAI: Grok 4.5 undercuts the market

• New model: Grok 4.5 launched after SpaceXAI’s acquisition of Anysphere (reported deal ~ $60 billion).
• Focus: Enterprise throughput and token compression, with launch pricing at about half the cost of some rivals.
• Performance: Independent tests show wins on speed and cost, while some competitors still lead on complex reasoning benchmarks. Expect pricing and performance pressure across the market.

Google: AlphaEvolve and broader frontier work

• AlphaEvolve: Testing a system that lets models rewrite their own code.
• Hardware & health: Continued investment in hardware and health-focused models like SensorFM, trained on large wearable datasets.
• Meta note: Meanwhile, Meta updated Muse Spark for workplace AI — big vendors are diversifying research and product offerings.

That’s it for now from Aurora and Isabelle.
Stay curious, stay skeptical — we’ll keep tracking the fast-moving AI world so you don’t have to.