PickAIModel Brief - Issue 005

Issue 005 — May 23, 2026 Independent AI model guidance: rankings, tools and industry analysis.

A note on format: we’ve dropped the “weekly” label. Issues will arrive when there’s enough happening to be worth your time — not because a calendar says we owe you an email. The content bar stays the same. The deadline pressure does not.

This issue is not a leaderboard recap.

The useful story this time is bigger than which model moved up a point. Google used I/O to turn Gemini into an agent platform, Anthropic plugged Claude into SpaceX-scale compute, and the AI business model is shifting from “chat with a model” to “give agents somewhere to run.” The rankings still matter — they’re what PickAIModel is built for — but the industry is now being shaped just as much by infrastructure, distribution, tools and trust.

Here’s what changed.

The Stories That Matter

Google I/O Was Not Just a Model Launch. It Was an Agent Stack Launch.

Google’s most important I/O announcement was not one model. It was the shape of the system around the models.

At Google I/O 2026, Google launched Gemini 3.5 Flash, introduced Gemini Omni for video-first creation, expanded Google Antigravity for developers, announced Gemini Spark as a persistent personal agent, and showed Universal Cart as an AI shopping layer across Search, Gemini, YouTube and Gmail. That is not a normal product refresh. It is Google trying to make Gemini the connective tissue across search, work, code, shopping and media.

The Flash release is still worth noting. Google says Gemini 3.5 Flash is generally available through Antigravity, the Gemini API in AI Studio and Android Studio, and that it outperforms Gemini 3.1 Pro on coding and agentic benchmarks such as Terminal-Bench 2.1, GDPval-AA and MCP Atlas. But the more important point is not the score. It is that Google is putting a cheaper, faster model at the centre of agent workflows.

Why it matters: If you build with AI, Google is no longer only selling “a model endpoint.” It is selling an execution environment: model, tools, agent runtime, Workspace access, shopping surface, media generation and cloud deployment. That is much harder to compare with a simple benchmark table.

Anthropic Plugged Claude Into SpaceX-Scale Compute

Anthropic’s SpaceX deal is the clearest reminder this month that the AI race is now an infrastructure race.

On May 6, Anthropic said it had signed an agreement with SpaceX to use all of the compute capacity at the Colossus 1 data centre, giving Claude access to more than 300 megawatts of new capacity and over 220,000 NVIDIA GPUs within the month. Anthropic said the added capacity would improve availability for Claude Pro and Claude Max subscribers, and it placed the deal alongside other large compute agreements with Amazon, Google, Broadcom, Microsoft, NVIDIA and Fluidstack. It also said it had expressed interest in working with SpaceX on orbital AI compute capacity. Anthropic announcement

That last detail sounds like science fiction until you look at the demand curve. Anthropic is not renting compute because it wants a better press release. It is renting compute because Claude usage has been running into visible limits.

Reuters later reported that Anthropic agreed to pay SpaceX $1.25 billion per month through May 2029 for compute services using SpaceX’s Colossus and Colossus II data centres, with either party able to terminate the arrangement on 90 days’ notice. Reuters

Why it matters: For normal users, “compute deal” sounds remote. It is not. It affects rate limits, latency, availability, price pressure and how aggressively a company can expand coding and agent features. Claude becoming more usable at peak times may matter more to paying customers than another abstract intelligence point.

Anthropic’s Revenue Is Now the Story Beneath the Valuation

The headline number around Anthropic is the possible valuation. The more useful number is revenue.

According to Reuters, Anthropic expects sales for the June quarter to exceed $10.9 billion, more than doubling from $4.8 billion in the March quarter, and is approaching its first quarterly operating profit. Reuters also reported expected Q2 operating profit of $559 million. Reuters

That is the part buyers should pay attention to. A profitable or near-profitable Anthropic has a different negotiating position from a lab burning cash while chasing the next model release. It can buy compute, hold product prices, invest in enterprise support, and absorb the expensive early years of agent adoption.

This does not mean Anthropic is “safe” or that every revenue projection will hold. AI infrastructure costs are still huge, and the SpaceX deal itself shows how much money is required just to keep up with demand. But the direction is clear: coding and agent automation are no longer a demo category. They are becoming a serious software revenue line.

Why it matters: If you are choosing a long-term AI platform for work, financial durability matters. The model has to be good, but the company also has to keep capacity online, support enterprise customers and survive the cost of serving heavy users.

The Benchmark Conversation Is Moving Toward Real Work

We are not turning this issue into a scoreboard. But one evaluation shift is important enough to mention.

The industry is moving away from treating one grand intelligence score as the whole story. Model launches now increasingly cite task-shaped evaluations: terminal work, multi-tool orchestration, agentic coding, browser use, finance work and long-context retrieval. That is a healthier direction. A model that wins a general reasoning index may still be the wrong choice for DevOps automation. A model that looks merely “good” on a broad intelligence index may be excellent at cheap, fast agent execution.

That is why PickAIModel will continue to separate the newsletter from the leaderboard. The leaderboard should carry the comparison burden. The Brief should explain the market context: which new tools are worth testing, which claims deserve scepticism, and what a buyer or developer should do next.

Why it matters: When a vendor says its model leads on a benchmark, ask whether that benchmark matches your workflow. “Best model” is becoming less useful than “best model for this task, at this price, with this reliability profile.”

Universal Cart Shows Where Consumer AI Is Heading

Google’s Universal Cart deserves more attention than it will probably get.

Google says Universal Cart will let users add products while browsing Search, chatting with Gemini, watching YouTube or reading Gmail. The cart then works in the background: finding deals, tracking price drops, showing price history and warning when an item comes back in stock. Google I/O announcements

That is not just a shopping feature. It is a distribution play. If AI assistants become the layer where people research, compare and eventually buy, then the battle for AI moves into commerce. Search ads, affiliate referrals, shopping comparison, product recommendations and checkout behaviour all start blending into the assistant experience.

There is an obvious trust problem here. A helpful shopping agent can save time. A conflicted shopping agent can quietly become an advertising interface wearing a helpful face. The difference will depend on disclosure, ranking transparency and how much control users get over recommendations.

Why it matters: AI shopping assistants may become one of the first mainstream places where ordinary users feel the trade-off between convenience and commercial influence. Watch how clearly Google labels paid placement, recommendation logic and price-history data.

Technology Shift of the Issue

Video Is Becoming a Native AI Input, Not a Stack of Screenshots

Gemini Omni is the clearest technical shift from Google I/O.

Google describes Gemini Omni as a model that can create “anything from any input,” starting with video. The first rollout focuses on video generation and editing through the Gemini app, Google Flow, YouTube Shorts Remix and YouTube Create. Google says Omni Flash can take image, text, video or audio references and turn them into a cohesive video output, with SynthID watermarking included.

The important change is not simply “better video generation.” It is the move toward video as a first-class medium for reasoning and editing. Most AI workflows still treat video awkwardly: sample frames, transcribe the audio, analyse the pieces separately, then stitch the answer together. Omni points toward a more natural workflow: ask the model to understand the scene, the motion, the voice, the style and the edit as one object.

Practical significance: Media teams should watch this closely, but so should training, sports analysis, surveillance review, education and support teams. Anywhere video carries the real context, native video models will matter.

Tool of the Issue

Google Antigravity

Agent-first development platform for building and managing AI agents

Google Antigravity is becoming the developer centre of Google’s agent strategy.

Google describes Antigravity as its agent-first development platform for taking an idea and turning it into a production-ready app. At I/O, Google expanded the Antigravity ecosystem across developer surfaces, while its developer site describes Antigravity as a standalone desktop app and CLI that lets builders work with agents that can operate autonomously in the background. Google developer highlights Google Developers

The practical appeal is obvious: developers do not want a dozen disconnected agent experiments. They want a place to define tasks, connect tools, observe runs, fix failures and deploy something usable. Antigravity is Google’s attempt to own that layer.

Best for: Developers already using Gemini, Google Cloud, Firebase, Android Studio or Workspace who want to test agentic coding and workflow automation without stitching together a custom orchestration stack.

Why this issue: Google I/O made Antigravity feel less like a side project and more like part of the main Gemini strategy.

Google Antigravity | Non-affiliate link

The Week in Accidental Honesty

The strangest sentence in AI this month is not a quote. It is the business arrangement itself: Anthropic is renting large-scale AI compute from SpaceX, while SpaceX is tied to Elon Musk’s own AI ambitions.

That is the whole market in one image. Everyone wants to beat everyone else. Everyone also needs GPUs, power and data-centre capacity badly enough to make uncomfortable partnerships look sensible.

On the Leaderboard

New frontier releases under review

`GPT-5.5` and `Gemini 3.5 Flash` are under review for the PickAIModel leaderboard.

They will not be ranked in the published roster until there is enough comparable evidence from accepted sources. Vendor launch claims are useful signals, but they are not enough on their own. We look for independent coverage, comparable benchmark conditions, price context, availability, and whether the model’s strengths show up in workflows buyers actually care about.

This issue intentionally avoids a full benchmark recap. The leaderboard will handle rankings. The Brief will focus on what the week’s AI developments mean.

What to Watch Next

• Gemini 3.5 Pro — Google says it is in testing and expected next month. The interesting question is not whether it scores higher than Flash, but whether it justifies the price and latency trade-off.
• Claude capacity after the SpaceX deal — users should watch whether rate limits, peak-time reliability and Claude Code availability noticeably improve.
• Agent platforms — Antigravity, Spark, OpenClaw, Bedrock AgentCore and similar systems are turning “AI model choice” into “AI operating environment choice.”
• AI commerce disclosure — Universal Cart may become an early test of how transparent AI shopping recommendations need to be.

If someone forwarded this to you, subscribe at pickaimodel.com/newsletter. Have a question for a future issue? Reply to this email — we read everything.

Disclosures

Editorial independence. PickAIModel.com produces independent editorial content. Model rankings, quality scores, and value scores are determined by our published methodology and are not influenced by commercial relationships with any AI vendor. No company can pay for ranking position, score changes, inclusion in rankings, or favourable treatment in our methodology outputs.

AI-assisted content disclosure. Portions of the editorial summaries and commentary in this newsletter may be drafted with the assistance of AI language models and reviewed by the PickAIModel editorial team. Any benchmark references in this issue are used as editorial context only. Official PickAIModel rankings are not invented by the model and are not editorially altered to favour advertisers or affiliates.

Affiliate disclosure. This newsletter may contain affiliate links. If you click a qualifying link and make a purchase, PickAIModel.com may earn a commission at no additional cost to you. Affiliate relationships do not influence our rankings, scores, or methodology outputs. The Tool of the Issue link in this issue is a non-affiliate link.

Not financial or legal advice. Nothing in this newsletter constitutes financial, investment, or legal advice. References to funding rounds, valuations, IPO timelines, pricing, contracts, or company strategy are provided for informational purposes only. Make decisions based on your own judgment and, where appropriate, qualified professional advice.

Accuracy and currency. AI model pricing, capabilities, availability and company claims change frequently. While we aim to be accurate at the time of publication, information may become outdated or be revised after publication. Verify critical purchasing details directly with the relevant vendor before acting.

To unsubscribe, use the unsubscribe link in your email footer. Copyright © 2026 PickAIModel.com. All rights reserved. NeuroSync AI Dynamics Pty Ltd.