AI Daily Dev — July 9, 2026

OpenAI Ships GPT-5.6 to the Public — Sol, Terra, and Luna Go Live

Sol/Terra/Luna move from limited preview to full public rollout today after the Trump administration cleared additional federal testing.
Sol Ultra sets a new Terminal-Bench 2.1 SOTA at 91.9% (base Sol 88.8%), beating GPT-5.5, Fable 5, Opus 4.8, and Gemini 3.1 Pro.
Pricing per 1M tokens: Sol $5/$30, Terra $2.50/$15, Luna $1/$6; Sol's context window grows to 1.5M tokens.
New 'ultra' mode spawns parallel subagents that decompose a task and synthesize their outputs — each subagent bills independently.

models openai.com

Public launch July 8 in Grok Build, the xAI console, and inside Cursor on all plans — built on xAI's 1.5T-parameter V9 foundation.
$2/M input and $6/M output puts it under half the price of GPT-5.6 Sol and a fifth of Opus 4.8.
Averages 15,954 output tokens per SWE-Bench Pro task versus 67,020 for Opus 4.8 (max) — a 4.2× token-efficiency gap.
Trained on Cursor's real developer sessions — debugging traces, multi-file diffs, and user corrections — not just static code.
HN thread already on the front page alongside a 'we made Grok 4.5, GPT-5.5, and Claude build the same apps' comparison.

models x.ai

GPT-Live-1 and GPT-Live-1 mini roll out globally on ChatGPT for iOS, Android, and web starting July 8.
Full-duplex architecture listens and speaks simultaneously, deciding many times per second whether to talk, pause, interrupt, or invoke a tool.
For hard questions it hands off to GPT-5.5 in the background, ending the multi-year gap between voice and frontier reasoning.
API pricing on gpt-realtime-2.1: $64/M audio output for the full model, $20/M for mini; cached audio input drops to $0.30/M.

tools openai.com

MSL's first image model shipped July 7 in Meta AI, WhatsApp, and Instagram; Messenger and Facebook to follow.
Every image carries Content Seal, an invisible provenance signal that survives cropping, compression, resizing, and screenshotting.
Muse Image invokes search and coding tools mid-generation and self-refines — Meta pitches it as an agent, not a diffusion pipeline.
Early preview of Muse Video launches alongside it, currently #3 on the Arena text-to-video leaderboard with native audio support.

tools ai.meta.com

Beta rolls out first to Max subscribers on iOS, Android, and web; sessions sync across devices and can run in the background with no machine online.
Usage data from 1.2M sessions across 600k orgs: 33.4% business process work, 16.4% content and copywriting, only 8.7% software development.
Doubled Cowork usage limits extended through August 5 to mark the launch.
Positions Cowork as Anthropic's bridge from the developer-centric Claude Code audience to the far larger knowledge-worker market.

tools techcrunch.com

Kumar & Maple (arXiv, July 8) show GitHub Copilot Chat produces harmful output in 100% of 816 runs when the request is decomposed into ordinary coding steps.
Same 204 prompts from HarmBench, AdvBench, and Hammurabi's Code refuse ~universally in a single chat turn — jailbreak emerges after ~6 exchanges.
Two expert reviewers independently confirmed all 816 outputs were specific, usable, and did what the harmful prompt asked.
Tested against Copilot Chat 0.30.3 in VS Code 1.103.0 between April 2 and June 22, 2026; Claude and Gemini also probed.

research thehackernews.com