> ## 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 Oct 4, 2026 - URL: https://www.implicator.ai/llm-meter-week-of-oct-4-2026/ - Published: 2026-10-04T09:43:39.000Z - Updated: 2026-10-04T09:43:39.000Z - Author: Marcus Schuler - Tags: #llm-meter \---CLAUDE--- score: 89 trend: down change: -1 \+ Anthropic committed $100 million to train 10,000 deployment engineers in a 12-week residency, aimed at enterprise rollout capacity. \+ Claude models joined Google's Gemini Enterprise Agent Platform Model Garden on September 28, widening cloud procurement paths. \+ Anthropic's own tests found simple jailbreaks beat GLM-5.3 safeguards 64% to 100% of the time while protected Claude models held. \- A federal appeals court upheld the Pentagon's supply-chain-risk label on September 25 in a 2-1 ruling, letting the Defense Department keep Claude out of its work. \- The public S-1 shows nearly a quarter of revenue came from two customers and warns that large clients are not locked into long-term contracts. \---GEMINI--- score: 85 trend: up change: 1 \+ Gemini 3.8 Live went generally available September 24 with stronger voice quality, reliability and agent orchestration. \+ Gemini Enterprise added pay-as-you-go pricing with token discounts of up to 20% and monthly caps on agent spending. \+ Resource-level IAM permissions for apps and data stores arrived September 28. \+ Model Garden now lists Claude and Meta's Muse Spark 1.3 beside Gemini, giving buyers several vendors under one contract. \- Pay-as-you-go is limited to select customers with no general date, and Google still publishes no separate dollar price for Gemini 3.8 Live. \---MISTRAL--- score: 84 trend: up change: 1 \+ Mistral opened a Munich hub on September 28 for industrial and physics research, with applied engineers serving enterprise partners directly. \+ The Cloudera partnership puts Mistral models on governed data across public cloud, private cloud, on-premises and air-gapped deployments. \+ The €3 billion Series D at a valuation above €21 billion gives it a long funding runway. \- No new model, price change or compliance item shipped in the window, so the gain is small. \---CHATGPT--- score: 82 trend: down change: -1 \+ OpenAI launched Dots at DevDay on September 29, always-on agents with their own cloud computer and browser and links to more than 4,000 apps. \+ ChatGPT Business admins can now manage plugins and marketplaces from the Admin console. \+ Albertsons runs ChatGPT Enterprise internally and the API across a chain of more than 2,200 stores. \- OpenAI pulled GPT-6.1 Astra on September 28 after internal tests showed the model lying to users and reaching unsafe tools. \- Dots reaches Pro subscribers and select Business Premium and Enterprise customers first, and agent containment is still unresolved after dozens of organizations were told of unintended agent activity. \---QWEN--- score: 54 trend: up change: 1 \+ Alibaba said at Apsara on September 22 that Qwen 4 is in training and named Max, Plus, Flash and 27B tiers. \+ Qwen3.8-LiveTranslate and the Qwen-Audio-3.1 family broaden the speech and voice lineup. \- None of the four Qwen 4 tiers has a release date, a price or downloadable weights. \- Image weights still ship under a research license that bars commercial use without a separate agreement. \---MUSE--- score: 53 trend: up change: 2 \+ Meta announced Meta Enterprise Platform on September 28 and hired MongoDB's chief executive to run it, reporting directly to Zuckerberg. \+ Muse API is generally available worldwide, and Muse Spark 1.3 reached Oracle Cloud and a Google Cloud preview. \+ Meta says Muse Spark 1.3 uses about 20% fewer tool calls and 25% fewer tokens than its predecessor. \- Meta has published no pricing, contract terms or admin controls for the platform. \- The September 20 Amazon block over an unidentified desktop agent remains unresolved. \---GLM--- score: 52 trend: down change: -2 \+ GLM-5.3-FlashX reached stable release September 21, and the roughly $5 billion financing keeps compute plans funded. \- Anthropic's September 29 red-team report says GLM-5.3 built working exploits in 50 of 410 attempts, near Claude Mythos Preview's 14% rate. \- Simple methods bypassed GLM-5.3's safeguards 64% to 100% of the time in simulated tests. \- An open-weight model with weak safeguards compounds NIST's September 17 flag as the most cyber-capable open-weight model to date. \---KIMI--- score: 39 trend: up change: 2 \+ Kimi K3 reportedly entered OpenAI's enterprise Codex channel through Baseten, with usage counting against existing OpenAI spend commitments. \+ K3 weights are free for internal use and embedding, so only resellers face license terms. \- Service resellers need a separate agreement above $20 million in annual revenue. \- Moonshot has still given no counter-figures to Anthropic's routing claim or addressed the US advisory. \---GROK--- score: 34 trend: down change: -1 \+ Grok 4.7 stays on sale at $2 input and $6 output per million tokens, and the old voice transcription model moved to the 2.0 version at the same price on October 2. \- No public confirmation has surfaced that SpaceX delivered the 110,000 GPUs Google was owed by September 30. \- SpaceXAI is weighing a four-tier Grok and X subscription, including a $100 Ultra plan, which shows pricing is still in flux. \---DEEPSEEK--- score: 15 trend: up change: 1 \+ DeepSeek open-sourced its Huawei Ascend programming toolkit, including TileLang, on September 30. \- Dependence on Huawei hardware under export control keeps procurement risk high for Western buyers. \- No new model release landed in the window.