Implicator
This week's LLM Popularity Meter records the widest score swings since its launch, with four models moving 4 points or more. Claude fell 5 points to 85% after two security incidents in five days. Mistral posted the biggest single-week gain, climbing 6 points to 61% on the strength of its first owned compute infrastructure.
Rankings held steady. Claude, ChatGPT, Gemini, Mistral, Grok, DeepSeek. But the gaps shifted. The spread between first and second narrowed to 5 points, the narrowest gap this year. And Grok's 6-point slide to 42% raises a question that matters if you track enterprise AI procurement: what counts as adoption when subscription purchases are mandated by contract?
What Changed This Week
- Claude drops 5 points to 85% after CMS leak and npm source code incident but holds first place
- Mistral posts biggest gain (+6 to 61%) with $830M debt raise for first owned European data center
- ChatGPT barely moves (+1) despite $122B funding as mobile and enterprise share erode
- Grok falls 6 points after NYT reports banks forced to buy subscriptions for SpaceX IPO access
AI-generated summary, reviewed by an editor. More on our AI guidelines.
Claude still leads, but the margin is thinner
Two incidents rattled enterprise confidence in the span of one work week. On March 26, roughly 3,000 internal files leaked through a CMS misconfiguration, exposing an unreleased model tier called Claude Mythos, described internally as having advanced cybersecurity capabilities. Five days later, 512,000 lines of Claude Code TypeScript source shipped accidentally via npm, revealing 44 hidden feature flags and an undisclosed stealth coding mode called Undercover. Anthropic filed more than 8,000 DMCA takedowns.
Then came the billing shock. Anthropic cut off OpenClaw subscribers on April 4, forcing heavy agent users from flat-rate plans to pay-as-you-go pricing. Some reportedly face costs up to 50 times higher.
Positive signals keep Claude from falling further. A new Compliance API gives financial services and healthcare clients programmatic audit log access for the first time. Microsoft 365 connectors expanded to free-tier users. The $400 million Coefficient Bio acquisition pushes Anthropic into life science AI. Trust took a hit. The fundamentals held.
Mistral made the week's most consequential move
The $830 million debt raise from a seven-bank consortium led by BNP Paribas funds 13,800 Nvidia GB300 GPUs for a 44-megawatt data center south of Paris. First owned compute. No more Azure, GCP, or CoreWeave dependency. For government clients who treated third-party cloud as a security disqualifier, the procurement math just changed.
Mistral also joined Nvidia's Nemotron Coalition to co-develop open frontier models on DGX Cloud. Combined with a $1.4 billion Sweden investment from February, the company is targeting 200 megawatts of European compute capacity by end of 2027. On track for $1 billion in annual recurring revenue.
But the gap persists. No US compliance certifications lock Mistral out of American enterprise procurement entirely. The 6-point rise to 61% still reflects something real, though. Sovereignty backed by owned infrastructure beats sovereignty as a pitch deck.
Get the weekly LLM Popularity Meter in your inbox
Strategic AI news from San Francisco. No hype, no "AI will change everything" throat clearing. Just what moved, who won, and why it matters. Daily at 6am PST.
No spam. Unsubscribe anytime.
ChatGPT and Gemini move for opposite reasons
OpenAI closed a $122 billion round at an $852 billion valuation. Largest private funding ever. Codex crossed $1 billion in annualized revenue as of January. Sora was shut down to redirect resources toward Codex and GPT-5.5.
ChatGPT moved 1 point. The reason sits in the other column. Mobile daily active user share fell below 40% for the first time, hitting 38.7% in March. Down from 52% just six months ago. Enterprise API share tells the same story. Roughly half the market in 2023. Closer to a quarter now. The COO shifted to a private equity role. The head of AGI deployment is on medical leave. The CMO stepped away for cancer treatment. Capital is keeping pace with the erosion. Not outrunning it.
Gemini gained 4 points on execution alone. Google released Gemma 4 under Apache 2.0 with four open-weight models, the 31-billion-parameter version ranking third on the Arena leaderboard and beating models 20 times its size. A five-tier API pricing structure gives enterprises granular cost control. Veo 3.1 video generation went free for all 3 billion Workspace users. Google Cloud Next on April 22 could push things higher.
Grok's enterprise numbers need an asterisk
The New York Times reported that Musk requires banks advising on the SpaceX IPO to purchase Grok enterprise subscriptions. Five of Wall Street's biggest advisory banks signed on, Morgan Stanley and Goldman Sachs among them. Some have already spent tens of millions. Contractual coercion is not organic demand. Those numbers deserve scrutiny.
All eight non-Musk xAI cofounders have now departed. Musk himself posted in March that xAI "was not built right first time around" and is being rebuilt from scratch. Zero compliance certifications. A 6-point drop to 42%.
DeepSeek approaches its floor
V4, expected to launch in weeks, will run exclusively on Huawei Ascend chips, according to The Information. Technically impressive. Strategically isolating. A 7-hour outage on March 30 hit 355 million users, the longest downtime since January 2025. NIST found DeepSeek agents 12 times more likely to follow malicious instructions than US frontier models. Government bans keep expanding across the US, Australia, Italy, and Taiwan.
At 15%, DeepSeek is close to the bottom. The remaining score reflects one audience: enterprises self-hosting open weights on their own infrastructure, sidestepping data sovereignty concerns entirely. That may be the only Western audience DeepSeek keeps.
Frequently Asked Questions
What is the LLM Popularity Meter?
A weekly editorial tool by Implicator.ai that scores six major AI models on enterprise satisfaction. Each model gets a percentage score based on adoption data, news coverage, and editorial analysis. Scores update every Saturday.
Why did Claude's score drop this week?
Two security incidents in five days damaged enterprise trust. A CMS misconfiguration leaked 3,000 internal files, then 512,000 lines of Claude Code source shipped via npm. Anthropic also cut off OpenClaw subscribers, raising costs for heavy agent users.
What drove Mistral's 6-point gain?
Mistral raised $830 million in debt to build its first owned data center near Paris with 13,800 Nvidia GPUs. Owning compute removes dependency on third-party cloud providers, which was a disqualifier for government clients requiring sovereign infrastructure.
Why did Grok drop despite new product features?
The New York Times reported Musk requires SpaceX IPO advisory banks to buy Grok subscriptions. This inflates enterprise adoption metrics through contractual coercion rather than organic demand. All eight non-Musk xAI cofounders have also departed.
How does DeepSeek's Huawei chip strategy affect its score?
DeepSeek V4 will run exclusively on Huawei Ascend chips, isolating it from Western enterprise buyers. Government bans continue expanding, and NIST found DeepSeek agents 12 times more likely to follow malicious instructions than US frontier models.
AI-generated summary, reviewed by an editor. More on our AI guidelines.
Marcus Schuler
Tech translator with German roots who fled to Silicon Valley chaos. Decodes startup noise from San Francisco. Launched implicator.ai to slice through AI's daily madness—crisp, clear, with Teutonic precision and sarcasm. E-Mail: [email protected]
