What Opus 5.5, GPT-6 Sol, and Meta's new AI glasses mean for you

Three frontier model launches, a $1,300 pair of AI glasses, and a keychain with a personality — all in the same seven-day stretch. If you blinked this week, you missed a genuine reset of the AI leaderboard, plus a hardware reveal that Meta has been teasing for years.

Here's the breakdown of what actually shipped, why it matters, and how to get your hands on it.

What made this week different from every other AI news cycle?

Most weeks in AI bring one notable release and a lot of noise around it. This week brought three separate frontier-model announcements from three different labs, a full hardware keynote from Meta, and a platform-wide feature drop from YouTube. Anthropic shipped a new Opus model, OpenAI pushed out its next-generation Sol tier, and Meta used its annual Connect event to finally reveal the tethered mixed-reality glasses it's been rumored to be building for years.

Layer on top of that the fact that OpenAI's Dev Day is only a week away, and you've got a leaderboard that could shift again before most people even finish reading about this week's changes.

Why is Claude Opus 5.5 such a big deal?

Anthropic quietly rolled out Claude Opus 5.5 as the newest member of its Opus lineup, positioning it as the strongest Opus model the company has shipped. According to Anthropic's own model page, Claude Opus 5.5 is Anthropic's strongest Opus model yet, powering long-running, highly capable agents while delivering improvements in coding and professional work, and it performs at the level of Claude Fable 5.1 on most work while costing 40% less to run than Opus 5.

That price drop matters just as much as the capability bump. Anthropic has been steadily making its top-tier models more affordable with each release — a pattern that started when Opus 4.5 cut pricing to $5/$25 per million tokens and made frontier-level intelligence accessible to far more developers. Opus 5.5 continues that trend, which is a big reason it's showing up as the default recommendation inside coding tools like Claude Code and third-party IDE integrations.

If you want to try it yourself, head to Claude's Opus model page or spin it up directly through the Claude Developer Platform if you're building agents or coding workflows.

What is GPT-6 Sol and how does it compare?

While Anthropic was updating Opus, OpenAI quietly rolled its GPT-5.6 lineup into a new generation. OpenAI's own model documentation now lists three tiers under the GPT-6 umbrella: GPT-6 Astra as the flagship model for complex reasoning and coding, GPT-6 Sol to balance intelligence and cost, and GPT-6 Luna for cost-sensitive, high-volume workloads. All three support text and image input, text output, multilingual capabilities, and vision.

That's a meaningful naming shift from the old numbered scheme. Instead of chasing decimal points (5.1, 5.2, 5.3...), OpenAI now separates models into durable capability tiers that can each get upgraded independently. Astra sits at the top for anyone doing heavy reasoning or agentic coding work, Sol is the balanced daily driver, and Luna handles cheap, high-volume tasks like classification or simple chat.

For coding specifically, GPT-6 Sol currently sits at the top of most benchmarks people are watching, edging out both Opus 5.5 and Grok in head-to-head coding tests — though as always with these leaderboard claims, the gap tends to shrink or flip within weeks as each lab ships incremental updates.

How do you access GPT-6 Sol and the new Agent API?

OpenAI paired this release with an expanded Agent API and a new "Live" mode built for real-time, voice-driven agent interactions. You can find the full technical breakdown in OpenAI's Agent and Live API guide, which walks through how to wire up persistent agents that can hold context across long-running conversations and tool calls.

To get started:

Head to developers.openai.com/api/docs/models and pick the tier that matches your use case — Astra for heavy reasoning, Sol for balanced everyday work, Luna for volume.
Review the Live API guide to understand how streaming and agent state work differently from a standard chat completion.
Test a small agent workflow in a sandbox environment before wiring it into production — especially if you're building anything that touches real user data or executes code autonomously.
Watch for updates from OpenAI's Dev Day, since pricing and rate limits on brand-new model tiers tend to shift quickly in the first few weeks after launch.

What did Meta announce at Connect 2026?

Meta used its Connect keynote to finally reveal the mixed-reality glasses that had been rumored under the codename "Project Phoenix" for over a year. The headline device, now officially called Meta VR Glasses, ships in spring 2027 and weighs about 100 grams on the face, uses a compute puck on an optical tether, and carries a price of $1,299.99.

The specs are genuinely impressive for something this light. The glasses run on 2.4K micro-OLED displays and are powered by a tethered puck with Qualcomm's Snapdragon Reality Elite chipset, and that puck does the heavy lifting — the glasses themselves have no onboard computing, not even a tracking coprocessor, relying entirely on the roughly 300-gram puck for both power and data over a thin cable.

Notably, there was no new standalone Quest headset this year — Meta is clearly betting its hardware roadmap on this lighter, glasses-first form factor instead.

Who is Jolly and what does Muse actually do?

Buried in the keynote was arguably the more interesting long-term bet: Meta's personal AI agent, Muse. Muse isn't brand new, but Zuckerberg used Connect to signal just how central it's about to become. As TechCrunch reported, Meta's personal AI agent Muse is only a few weeks old, and CEO Mark Zuckerberg kicked off Connect with a keynote that made clear the company is going all in on it. The agent even got a name for its physical avatar — the Muse agent's physical avatar is named Jolly, and it's designed to handle everyday tasks by connecting to users' apps and services, including email and calendars.

Meta didn't stop at software either. Toward the end of the keynote, the company teased a dedicated hardware companion for the assistant — a compact keychain device called Muse Charm, equipped with a small display, speakers, microphones, and a fingerprint sensor, delivering real-time voice and avatar interaction in a portable form.

This is worth watching closely. Google has Gemini baked into everything, OpenAI has ChatGPT, and now Meta is trying to make Muse the connective tissue across its glasses, headset, and phone apps all at once.

What changed for YouTube creators and viewers?

The same week also brought YouTube's annual Made on YouTube event, and the AI-driven updates here are aimed squarely at both sides of the platform — people watching and people posting. For viewers, the standout feature is Custom Feeds, which lets you curate your homepage based on specific moods or interests, alongside a new conversational search tool called Ask YouTube.

For anyone who listens to music or podcasts on the platform, YouTube also rolled out Ask Music, a conversational AI tool that works across podcasts and a catalog of more than 300 million songs, letting users customize their queue and explore more information about artists. It's paired with Podcast Lineup, described as a personalized AI-audio guide that provides a weekly spoken preview of recommended shows, explains why they're being recommended, and takes users to suggested episodes.

Creators get their own upgrade too — a conversational, Gemini-powered editing assistant baked directly into Shorts and the YouTube Create app, letting you trim clips, sync music, and reorder footage using plain-language prompts instead of manually scrubbing a timeline.

Which model should you actually use right now?

If you're picking a model today, the honest answer depends entirely on what you're doing:

Heavy coding or long agentic workflows: GPT-6 Astra or Opus 5.5 are both strong picks right now, with GPT-6 Sol currently benchmarking ahead on several coding tests.
Cost-sensitive, high-volume tasks: GPT-6 Luna or one of Anthropic's smaller Claude tiers will save you real money without a huge capability hit.
Real-time voice agents: OpenAI's new Live API is purpose-built for this and worth testing over older chat-completion-based approaches.

Whatever you pick today, don't get too attached. With OpenAI's Dev Day happening just days after this news cycle, there's a real chance the leaderboard order changes again before you've finished reading your second cup of coffee.

What should you actually do with this information?

Don't chase every headline. Pick the model tier that matches your actual workload, bookmark the OpenAI models documentation and Anthropic's Opus page so you can check pricing when it inevitably shifts again, and keep an eye on Meta's glasses roadmap if you're curious about where consumer AI hardware is heading next. The pace of releases isn't slowing down — the only real strategy is picking your tools deliberately instead of reactively.