GPT-5.6, AI Search Chaos, and the Data Rights Backlash

James Turner

So, fifty-three point six. That is the number that should be keeping every developer and product manager awake tonight. It's the score Sol, OpenAI's new flagship model under the GPT-5.6 umbrella, just put up on the Agents' Last Exam benchmark. To put that in perspective, Claude Fable 5 scored a forty-eight point two. We are talking about a massive, massive leap in raw agentic capability, and it's officially rolling out to the public right now alongside its smaller siblings, Terra and Luna. And the performance? Oh my god. Cerebras is already running this thing at seven hundred and fifty tokens per second. It is-it is just blistering, instantaneous output. But as a developer, what really gets me is the pricing. OpenAI is positioning these perfectly. Sol is coming in at five dollars per million input tokens and thirty dollars per million output. Terra, the mid-tier workhorse, is at two-fifty and fifteen dollars. And Luna, the lightweight edge model, is a dirt-cheap dollar and six dollars. They are-they are clearly aiming to lock down the enterprise market before anyone else can even breathe.

James Turner

But, uh, the real story here isn't just the raw speed or the benchmarks. It's the absolute political firestorm happening behind the scenes. OpenAI has been very, very loud about their weeks of coordination with the US Commerce Department's AI Safety Institute, framing this rollout as this highly responsible, coordinated effort. They're basically saying, "Look, we are-we are building trust with the federal government." But then, boom. A White House official completely blindsides them, publicly disputing the whole narrative. They-they straight-up said no formal pre-clearance or federal approval was ever required or, more importantly, granted under current voluntary policies. It's this beautiful, tense game of regulatory theater. OpenAI wants the market to think they have the government's stamp of approval, while the White House is desperately trying to clarify that they aren't rubber-stamping proprietary systems.

James Turner

And honestly, when you look at what Sol can actually do, you see why everyone is sweating. It features this "ultra" mode that orchestrates four subagents running in parallel, spinning up their own code environments to solve complex, multi-step engineering tasks. It's incredibly powerful, but it also introduces serious, serious risks. OpenAI actually had to build a hardware-level gate for it. If you want to run Sol's advanced cyber or biochemical workflows, you literally have to authenticate with physical hardware passkeys under their new "Trusted Access" framework. We are-we are officially past the era of simple API keys, and entering the era of cryptographic physical security just to run code.

James Turner

Now, this whole shift in how we interact with these models is bleeding directly into how we search. Google just dropped some fascinating data from their first full year of "AI Mode." The way we search is fundamentally breaking. User search prompts are now three times longer on average than they were last year, and follow-up queries are growing at forty percent month-over-month. People are-they're actually conversing with the search engine, treating it like an assistant rather than just dumping keywords into a blank box. But there's a brutal flip side to this. A new randomized study just revealed that Google's AI Overviews are slashing organic traffic to external websites by thirty-nine point eight percent. Nearly forty percent of traffic, gone. And what hurts the most is that this isn't just low-value informational clicks. AI Overviews are capturing the most highly engaged, high-intent prospects before they ever reach a publisher's site. If you're building an SEO-dependent business, the ground is literally dissolving under your feet.

James Turner

And that brings us to the inevitable, massive backlash over data rights that has completely boiled over in the last forty-eight hours. Companies are desperate for training data, and users are finally pushing back hard. Look at HubSpot. They tried to quietly update their terms to use customer CRM data for AI features, and the community absolutely erupted. The backlash was so fast and so severe that HubSpot had to do a massive, embarrassing retreat, reversing the policy almost immediately. But then you have Meta, who is just-they are playing by entirely different rules. They've rolled out a highly controversial default opt-out policy for scraping public Instagram photos to train their image generators. If you don't want your personal photos used, you have to dig through buried privacy settings and manually request to opt out. It's a classic Meta move, and it's setting up a massive legal showdown over who actually owns the content we put online.

James Turner

It is wild out there. From hardware-locked agentic APIs to the slow death of traditional SEO, the landscape is shifting daily. We'll keep tracking it. Catch you in the next one.