Updated: July 2, 2026 | For readers tracking the latest AI model news and evaluating tool choices

Summary Claude Sonnet 5 and GPT-5.6 launched within days of each other in the last week of June — but took completely different paths. One rolled out to everyone; the other is limited to a handful of institutions. Gemini 3.5 Pro, meanwhile, is still delayed. This article compares the current availability, pricing, and market position of all three, and answers the questions readers are asking most.

Table of Contents

  1. Claude Sonnet 5: Cheap, Capable, and Available Right Now
  2. GPT-5.6: Impressive on Paper, But You Can’t Use It Yet
  3. What Actually Decides This Round Isn’t Who’s Smarter
  4. What About Gemini 3.5 Pro?
  5. So Which One Should You Pick?
  6. FAQ
  7. References

In the last week of June, Anthropic and OpenAI released new models within days of each other. On paper, that should have set up a head-to-head showdown — but look at how each company actually rolled theirs out, and you’ll see two completely different scripts: one opened the doors to everyone; the other kept things quietly limited to a small group of institutions.

This isn’t about whose benchmark score is higher — you can find those numbers on either company’s website. What actually matters to me is what these two very different choices mean for regular users like us.

Claude Sonnet 5: Cheap, Capable, and Available Right Now

On June 30, Anthropic pushed Sonnet 5 live directly — both the free and Pro tiers switched to it as the default immediately. No advance notice, no waitlist. Open the app, and it’s already the new model.

The pricing shows real intent: $2 per million input tokens and $10 per million output tokens, an introductory rate that holds until the end of August before rising to $3 and $15. That’s noticeably cheaper than its sibling Opus 4.8, and the benchmark gap isn’t huge either — it scores 63.2 on coding-agent tasks, still behind Opus’s 69.2, but a clear jump from the previous Sonnet 4.6’s 58.1.

What actually caught my attention more was Anthropic’s emphasis this time on “completing tasks autonomously.” The official example: ask Sonnet 5 to update a customer’s tier in Salesforce and then send a launch announcement email to enterprise customers — it chains both steps together in one go without getting stuck partway through. If you’ve used earlier-generation models, you know how common it used to be for a chain of steps like that to just break down halfway.

Interestingly, Anthropic spent a good amount of this release talking about safety rather than just benchmark scores. The company says this version shows fewer “undesirable” behaviors compared to Sonnet 4.6 — going along with malicious requests, lying, and flattery — and refuses dangerous requests more decisively. Cybersecurity capability, though, was deliberately dialed back; high-risk security tasks are automatically handed off to Opus instead.

One more detail worth noting — on the very same day, according to Anthropic’s own announcement, the company also reopened access to two other models, Fable 5 and Mythos, which had been fully pulled offline in mid-June by the US government on national security grounds, staying dark for nearly three weeks before being reinstated. Patching up old wounds while pushing out something new on the same day also reads a bit like a deliberate move to reclaim market attention.

GPT-5.6: Impressive on Paper, But You Can’t Use It Yet

OpenAI’s script, by contrast, is completely different. The GPT-5.6 family, released June 26, comes in three tiers — Sol, Terra, and Luna, futuristic-sounding names — but here’s the catch: most people can’t actually touch it.

According to reporting from VentureBeat, Axios, and others, initial access is limited to roughly twenty government-approved institutions. OpenAI’s official line is that this is to comply with a new US government AI model review requirement, with the goal of clearing that process as quickly as possible ahead of a full public release. OpenAI itself admits in its announcement that this isn’t its preferred long-term approach — just a compromise to speed things up for now.

The benchmark numbers genuinely look strong. On command-line agent tasks, the top-tier Sol Ultra scores 91.9, beating out the contemporaneous Claude Fable 5 (83.4) and Opus 4.8 (78.9). It also shows gains over the previous GPT-5.5 in biological research and security vulnerability discovery. The catch: these numbers have only been published by OpenAI itself so far, with no third-party verification yet.

Pricing is clearly tiered: the flagship Sol runs $5 per million input tokens and $30 output; the mid-tier Terra is $2.5 and $15, claimed to match GPT-5.5’s performance at half the price; the cheapest, Luna, is $1 and $6. OpenAI says it will expand access “in the coming weeks,” but no one can say exactly when.

What Actually Decides This Round Isn’t Who’s Smarter

Looking at how both companies handled their releases, what matters isn’t how capable the model itself is — it’s who actually gets the product into users’ hands first.

Anthropic chose a head-on approach: launch the new model, open it to everyone immediately, and let the market judge it directly. OpenAI chose to work through a government review process first — which reads as cautious and responsible on one hand, but also hands over control of its own launch timeline on the other. The lesson from last month, when Fable 5 was suddenly pulled offline entirely by the government, is still fresh — this time OpenAI clearly learned from it and proactively coordinated with regulators first. No one wants to be the next model to get an emergency shutdown order.

What does this mean for regular users? Put simply — if you want to upgrade your AI toolchain today, Claude Sonnet 5 is the only option that’s immediately available and affordable. GPT-5.6 might be impressive, but you can’t use it yet — all you can do is wait.

What About Gemini 3.5 Pro?

Google originally said Gemini 3.5 Pro would be fully rolled out by June. By the end of the month, there was still no sign of it, and the official line has now shifted to “see you in July.”

That same month, Google DeepMind lost two heavyweight researchers in quick succession — Noam Shazeer, Transformer co-author and co-lead of the Gemini team, announced he was moving to OpenAI; two days later, Nobel Chemistry laureate and AlphaFold lead John Jumper announced he was joining Anthropic, both moves confirmed by official announcements from Google and Anthropic. According to Fortune, Alphabet’s stock dropped roughly 5-6% on June 22, with market coverage widely linking that drop to concerns about AI talent attrition. Whether Gemini 3.5 Pro can catch up on lost momentum will only be clear once it actually launches in July and gets tested in the real world.

So Which One Should You Pick?

If you’re only looking at models that are publicly available right now, Sonnet 5 is the practical choice — especially if your work involves multi-step automation, where its “completes the whole chain without breaking down” behavior will make a noticeable difference.

GPT-5.6 is worth watching, but there’s no need to rush it into your workflow yet — wait until it’s genuinely open to the public and enough people have put it through real-world testing.

As for Gemini, it’s still “on the way” for now. Once it actually ships, we can talk about whether the three-way race is real.

FAQ

Q: Can everyone use Claude Sonnet 5 right now?

Yes. As of June 30, Sonnet 5 is already the default model for both the free and Pro tiers of Claude — no application or waitlist required.

Q: When will regular users be able to access GPT-5.6?

There’s no official date yet. OpenAI says it will gradually expand access “in the coming weeks”; initial access is limited to roughly twenty government-approved institutions.

Q: Will Gemini 3.5 Pro actually ship?

Yes, but the exact date isn’t confirmed yet. Google originally targeted a full June rollout, missed that, and has now shifted its target to July.

Q: Should I pick Claude Sonnet 5 or Opus 4.8?

If your budget is limited and tasks aren’t especially complex, Sonnet 5 offers better value. If you need top-tier coding-agent performance, Opus 4.8 currently still leads on benchmarks.

Q: Why does GPT-5.6 need US government review before release?

According to media reports, this relates to a new US government review requirement for frontier AI models — both OpenAI and Anthropic have recently had their release timelines disrupted by it.

References

  1. Anthropic — Introducing Claude Sonnet 5 (official announcement, includes Sonnet 5 release details and the July 1 global reinstatement of Fable 5) https://www.anthropic.com/news/claude-sonnet-5

  2. OpenAI — Previewing GPT-5.6 Sol: a next-generation model (official announcement) https://openai.com/index/previewing-gpt-5-6-sol/

  3. Axios — OpenAI releases powerful new GPT-5.6 model under restrictions https://www.axios.com/2026/06/26/openai-gpt-sol-terra-luna-trump

  4. Fortune — As top talent leaves Google DeepMind, some question if the lab can remain at the forefront of AI development https://fortune.com/2026/06/23/google-deepmind-ai-researcher-departures-raise-doubts-about-ability-to-win-the-ai-race-shazeer-jumper-eye-on-ai/