AI lab product releases reached a relentless pace this week, with OpenAI, Meta, SpaceXAI, and Anthropic all unveiling new models, features, and personnel changes across roughly 72 hours, leaving little doubt that any notion of a summer slowdown has evaporated.
OpenAI’s GPT-5.6 Family Sets a New Coding Benchmark
On Wednesday, OpenAI debuted GPT-Live, a voice model designed to make ChatGPT Voice feel closer to natural conversation, including the ability to speak over it. The following day, the company launched the GPT-5.6 family: Sol, the flagship reasoning model; Terra, aimed at everyday use; and Luna, a cost-saving option.
The performance claims are striking on paper. According to OpenAI’s product page, Sol scores 73.5% on ExploitBench, a benchmark measuring progress from reaching vulnerable code through to arbitrary code execution, against GPT-5.5’s 47.9% at a comparable output-token budget. On ExploitGym, which asks agents to convert real-world vulnerabilities into working exploits, Sol nearly doubles GPT-5.5’s peak pass rate, rising from 15.1% to 24.9% under a two-hour cap and reaching 33.7% with a six-hour cap.
OpenAI also says Sol with max reasoning performs within one point of Anthropic‘s Fable 5 on the Artificial Analysis Intelligence Index while completing tasks in 61% less time at roughly half the estimated cost.
According to 9to5Mac, the models arrived approximately two months after GPT-5.5, initially rolling out to a select group of trusted partners before general availability. The company also introduced new ‘max’ and ‘ultra’ reasoning effort options, stating: ‘With GPT-5.6, we’re introducing a new max reasoning effort to give Sol the most time to reason deeply.’
Sol, Terra, and Luna are available in ChatGPT, Codex, and the API, with Terra and Luna specifically accessible through ChatGPT Work, which OpenAI announced as a new workplace-oriented agent. The company also integrated Codex into the ChatGPT desktop app and confirmed it is sunsetting its Atlas browser as it narrows its focus away from peripheral projects.
On the personnel front, OpenAI’s chief futurist departed after nine years, and Fidji Simo, the company’s CEO of applications, stepped down following medical leave.
A Week of AI Lab Product Releases: Meta’s Muse and the Instagram Controversy
Meta’s superintelligence unit, led by Alexandr Wang, has been releasing products at pace since breaking cover with Muse Spark in April. This week the company debuted Muse Image, a new image-generation model that immediately drew criticism over its default privacy settings.
According to Wired, Instagram’s help centre states that ‘people may be able to create content with your Instagram content using AI features at Meta’ if users leave their account public and on default settings, and that opting out only prevents future AI-generated images while existing images remain in Meta’s possession. Business Insider reports the rollout is part of Meta’s broader push to compete with image-generation tools from OpenAI, Google, Midjourney, and Adobe by making AI image creation a built-in feature for Instagram’s billions of users.
Meta also released Muse Spark 1.1, the next iteration of its Muse model and the first the company has charged for. Chief executive Mark Zuckerberg described its pricing as ‘very low’ while criticising rivals for charging at ‘very extreme’ levels.
SpaceXAI released Grok 4.5 on Wednesday. Elon Musk described it as an ‘Opus-class model’ and said he was ‘clearly wrong about Anthropic,’ calling it the clear AI leader. Anthropic has been paying SpaceX $1.25 billion a month for compute. Grok 4.5 is the first model SpaceXAI has released since its acquisition of Cursor.
Anthropic, for its part, launched a ‘Reflect’ feature inside Claude, allowing users to see a summary and analysis of their habits with the model. Separately, TechCrunch reports that Anthropic’s $1.5 billion copyright settlement, believed to be the largest in the history of US copyright law, received final court approval this week, though many authors and creators still do not view it as a win.
Behind the scenes, Perplexity is also quietly developing a coding tool called Teammate, aimed at competing with Anthropic and OpenAI in that space, according to Business Insider.
Investor Jason Calacanis put it plainly on X: ‘It’s like a better new operating system, laptop and CPU being launched every 14 days.’ With AI lab product releases showing no sign of slowing, the next wave of announcements could arrive before the week is out.