> ## Content Index
> Fetch the complete content index at: https://www.implicator.ai/llms.txt
> Use this file to discover other available public pages before exploring further.

# LLM Meter — Week of May 10
- URL: https://www.implicator.ai/llm-meter-week-of-may-10/
- Published: 2026-05-10T13:31:38.000Z
- Updated: 2026-05-10T13:31:38.000Z
- Author: Marcus Schuler
- Tags: #llm-meter

\---CLAUDE---  
score: 89  
trend: up  
change: +2  
\+ $1.5B Wall Street venture with Goldman, Blackstone, Hellman & Friedman + Apollo, GIC, Sequoia — direct shot at McKinsey/BCG/Bain for the corporate AI transformation budget  
\+ Ten Claude finance agents launched May 5 with full Microsoft 365 integration GA, Moody's native app on 600M companies, plus Verisk, Third Bridge, D&B, Experian, GLG joining the data-partner roster  
\+ SpaceX deal locks all of Colossus 1 — 300+ MW and 220,000+ Nvidia GPUs (H100/H200/GB200) — coming online within the month, plus disclosed interest in orbital compute  
\+ Claude Code limits doubled and peak-hour throttling removed for Pro/Max, Opus API rate limits raised — directly addresses last quarter's renewal-cycle capacity friction  
\- Anthropic excluded from May 1 Pentagon IL6/IL7 classified-network vendor list of seven; supply-chain risk designation and Hegseth transition order still unresolved  
\---GEMINI---  
score: 87  
trend: up  
change: +1  
\+ Gemini Enterprise Agent Platform launched at Cloud Next '26 — Agent Designer, Studio, Registry, Identity, Gateway, Observability under one procurement umbrella  
\+ Pentagon cleared Google for IL6/IL7 classified networks May 1 alongside OpenAI, Microsoft, AWS, Nvidia, SpaceX, Reflection  
\+ GenAI.mil hit 100,000+ AI agents built and 1.3M active users on the IL5 platform after Gemini 3.1 Pro moved from preview to GA  
\+ $750M innovation fund for partner-built agents formalizes the channel-incentive layer beneath the platform  
\- 600+ DeepMind employee letter to Pichai opposing classified Pentagon work remains unresolved internally; some Cloud Next announcements were already absorbed in prior weeks  
\---CHATGPT---  
score: 83  
trend: up  
change: +1  
\+ The Deployment Company JV finalized May 4 — $10B with TPG, Brookfield, Advent, Bain Capital, Goanna; $4B+ raised, 19 investors, mirrors Anthropic's distribution play  
\+ GPT-5.5 API revenue grew 2x+ faster than any prior launch in week one; Codex revenue doubled in seven days; Terminal-Bench 2.0 at 82.7% leads the coding benchmark  
\+ OpenAI cleared for Pentagon IL6/IL7 classified networks May 1, restoring federal-tier procurement parity  
\- 17.5% guaranteed annual return to JV investors implies a premium cost of capital for distribution versus Anthropic's pure-equity Wall Street syndicate  
\- Q4 2026 IPO timeline now public; \~$14B projected 2026 losses and the $600B compute-commitment stack remain the structural overhang  
\---MISTRAL---  
score: 74  
trend: up  
change: +1  
\+ Mistral Medium 3.5 (128B dense, 256K context) launched May 2 with Vibe Remote Agents — cloud coding agents with GitHub, Linear, Jira, Sentry, Slack, Teams integration  
\+ 77.6% on SWE-Bench Verified — competitive agentic-developer benchmark, on top of the Workflows orchestration engine launched April 28  
\+ Le Chat Work mode runs on Medium 3.5 for parallel tool-calling, unifying the consumer and developer agent runtime under one procurement object  
\- Coding benchmarks still trail GPT-5.5 Terminal-Bench and Claude Opus 4.7 on parts of the stack  
\- Not on the Pentagon IL6/IL7 vendor list — structurally appropriate for a French firm but caps the federal-tier ceiling on this scoreboard  
\---GROK---  
score: 33  
trend: down  
change: -2  
\- Grok/xAI excluded from May 1 Pentagon IL6/IL7 classified-network roster despite SpaceX being on it — federal-tier signal that the Musk-aligned defense procurement track does not see Grok as the federal model  
\- SpaceX rented all Colossus 1 AI capacity to Anthropic on May 6 — xAI's most-touted infrastructure narrative now powers a direct competitor  
\- Musk-Altman trial coverage continues with the Apr 30 distillation admission still circulating; April 21 outage tail mostly resolved but news cycle has not closed  
\+ Grok 4.3 full rollout expected mid-to-late May with native video input, file output, 1M token context, $1.25/$2.50 per M tokens  
\+ Cursor compute-sharing commitment remains a distribution wedge if product traction follows  
\---DEEPSEEK---  
score: 14  
trend: down  
change: -1  
\- Canada CSE penetration test (May 2026) extracted system-level instructions and API keys, bypassed moderation filters via phonetic obfuscation specific to DeepSeek's tokenizer  
\- DeepSeek refused Canada Digital Government's request for internal red-team reports on commercial-confidentiality grounds, converting a security finding into a procurement-process finding  
\- Pentagon May 1 IL6/IL7 vendor list contains zero Chinese model providers; EU, Canada, South Korea, Australia, India restrictions remain in force  
\+ V4-Pro at $1.74/$3.48 per million tokens still roughly one-seventh of GPT-5.5 output cost; MIT license preserves on-premises optionality  
\+ Huawei Ascend and Cambricon serving stack is the structural APAC distribution story Western compliance windows cannot reach