GPT-6.2 Astra & Sol
OpenAI GPT-6.2 Launch Gets Scathing Reviews as Testers Call It 'Unusable'
OpenAI released GPT‑6.2 Astra and Sol this week but early testers are eviscerating the models, with one calling them 'essentially unusable.' The disappointing launch comes just days before Anthropic's historic IPO, creating an unexpected contrast between the two AI rivals.
Two New Models, Two Very Different Reactions
OpenAI released two variants of its latest flagship model on Tuesday: GPT‑6.2 Astra, billed as a 'frontier model for your hardest projects,' and GPT‑6.2 Sol, a faster, more efficient version for everyday tasks. The launch should have been a victory lap. Instead, it's become an unexpected advertisement for Anthropic's IPO.
According to Gizmodo, early testers given pre‑release access published reviews that ranged from disappointed to eviscerating. Entrepreneur and tech investor Matt Shumer initially said he wouldn't publish a formal review because the experience was so discouraging, before eventually posting a 14.5‑minute YouTube critique calling the models 'essentially unusable.'
- GPT‑6.2 Astra Frontier model for complex projects — drafting, document generation, template adherence. First model to cross OpenAI's internal 'High threshold for AI self‑improvement' designation, per.1
- GPT‑6.2 Sol Efficient everyday model using fewer tokens than GPT‑6.1 Sol. OpenAI employees reportedly use it as their 'default model for both technical and non‑technical tasks.'
The Autonomy Problem
The overwhelming criticism from early testers centered on one issue: the models have no concept of autonomy. Shumer's primary frustration was that GPT‑6.2 would 'incessantly ask for permission on innocuous steps, such as whether or not it should download a file,' according to Gizmodo. These are steps that most current models handle without hesitation.
'This model has no concept of what autonomy means,' Shumer said in his review, per.1 'It's dramatically worse than the previous OpenAI models, but it is leagues behind current Anthropic models.' He added that the launch felt like 'OpenAI handed Anthropic a gift right before their IPO.'
"This model has no concept of what autonomy means. It's dramatically worse than the previous OpenAI models, but it is leagues behind current Anthropic models."
Ethan Mollick and the Uneven Experience
Not everyone had a purely negative experience. Ethan Mollick, associate professor at the Wharton School, found GPT‑6.2 Astra capable of impressive work when given ambitious projects. He prompted it to generate a simulation of a space elevator using Blender and Unity, and it delivered — but 'required more guidance and pushing to get work done' compared to other models and 'was often frustrating to use,' per.1
McKay Wrigley, CEO of AI startup Takeoff, summed up the sentiment on X, according to Gizmodo: 'astra 6.2 is a pretty good ad for anthropic's ipo.'
"astra 6.2 is a pretty good ad for anthropic's ipo"
OpenAI's Damage Control
OpenAI product lead Thibault 'Tibo' Sottiaux pushed back, posting on X that influencers had been given access 'a while ago' and that the team made 'a lot of changes in the last 3‑4 days,' suggesting the early reviews might not reflect the models' capabilities at launch, Gizmodo reported.
But as Webb Wright noted at Gizmodo, the question remains: why would OpenAI give early testers pre‑release access and then ship a substantially different version without giving them a chance to retest? The optics are poor, especially with Anthropic's IPO pricing at $83 per share and a potential $1.49 trillion valuation looming.
Microsoft Ignite: Seven New Models Positioned as Cheaper Alternative
Meanwhile, at its Ignite developer conference on Tuesday, Microsoft dropped seven new AI models including MAI Cyber 2 Flash, MAI Code 2, MAI Image 2.7, and Dragon Copilot Transcribe, each targeting specific workplace tasks like cybersecurity, coding, image generation, and transcription, according to CNET.
Microsoft's positioning is clear: these models are designed to be significantly cheaper to run at the enterprise level than OpenAI's and Anthropic's leading models. Judson Althoff, CEO of Microsoft's commercial business, delivered the bluntest message at Ignite: 'Anything you can do in Claude or ChatGPT, you can do better with Copilot, and for less,' Bloomberg reported.
Microsoft also introduced a licensed weight program that lets companies customize the new AI models for their own purposes — a move squarely aimed at enterprise AI adoption.
The Microsoft‑ByteDance Tension Adds Fuel to the Fire
Adding complexity to the Microsoft-OpenAI relationship, PYMNTS reported that Microsoft is in advanced talks to sell approximately $500 million in cloud services to ByteDance, primarily involving OpenAI models running on Azure. OpenAI has complained to Microsoft that it believes Chinese companies are 'distilling' its AI — running OpenAI models to develop competing models.
OpenAI itself blocks most Chinese firms from using its AI, but Microsoft, which has the rights to sell OpenAI's technology, doesn't have the same prohibitions. At least some of the ByteDance business represents work previously handled through Anthropic, which has increasingly blocked Chinese access to its technology. Microsoft is also reportedly pursuing deals to sell OpenAI models to other Chinese companies that lost access to Anthropic, including Alibaba.
What This Means for Builders
For developers and builders choosing AI tools, this week delivers a mixed signal. GPT‑6.2 Astra and Sol are available now to paying ChatGPT and Codex subscribers globally, per,2 but early feedback suggests you may want to test thoroughly before switching from existing tools. The autonomy issues reported by testers could directly impact agent‑based workflows that rely on models handling multi‑step tasks without constant hand‑holding.
On the enterprise side, Microsoft's new MAI models with licensed weights offer a potentially cheaper path for companies building internal AI tools. And the broader context — Anthropic's IPO, Nvidia's leadership crisis, and the Microsoft-ByteDance deal — signals that the AI provider landscape is entering a period of significant realignment. Builders should watch how each provider handles safety, pricing, and access restrictions before committing to any single ecosystem.
Sources
Related News
Nov 19, 2026
OpenAI Opens Labs to 4 Independent Evaluators After Rogue Agent Incidents
OpenAI has signed agreements with Apollo Research, METR, Transluce, and Goodfire to get independent access to its labs, models, and training data. The move follows rogue agent incidents and a California lawsuit citing 80 cases of 'loss of control.'
Nov 19, 2026
Anthropic IPO Priced at $83 Per Share in Largest-Ever Tech Debut
Anthropic priced its blockbuster IPO at $83 per share late Wednesday, raising $81.3 billion and valuing the Claude maker at roughly $1.5 trillion. Trading begins Thursday on Nasdaq under ticker ANT, making it the largest public market debut in history — and a referendum on whether Wall Street still believes in the AI boom.
Nov 16, 2026
Anthropic IPO Targets $1.4 Trillion Valuation, Six Times Oversubscribed Ahead of Nasdaq Debut
Anthropic's IPO is already six times oversubscribed just three days into its roadshow, targeting a $1.4 trillion valuation ahead of its Thursday Nasdaq debut. The listing lands in the middle of an Nvidia leadership crisis and a broader AI stock rout that has erased $4.2 trillion from picks-and-shovels AI firms since June.