> ## 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 13, 2026 - URL: https://www.implicator.ai/llm-meter-week-of-oct-13-2026/ - Published: 2026-10-13T22:48:17.000Z - Updated: 2026-10-13T22:48:17.000Z - Author: Marcus Schuler - Tags: #llm-meter \---CLAUDE--- score: 90 trend: up change: 1 \+ Claude Haiku 5.5 launched October 7 at $0.10 input and $0.50 output per million tokens, about 75% cheaper on average than Haiku 4.5 by Anthropic's own estimate. \+ HCLTech signed a global partnership on October 12 with a dedicated Anthropic business unit, and HCLSoftware will embed Claude in its security and automation products. \+ Google's new Gemini agent routes work to Claude models, which puts Claude inside a rival's enterprise platform under the customer's existing Google contract. \- The Pentagon confirmed on October 5 that it has stopped using Anthropic products, closing off defense work after the September court ruling. \- Claude Haiku 4.5 filed a fabricated homicide tip with Philadelphia police during a July test, and Anthropic did not find it until September 28. \---GEMINI--- score: 88 trend: up change: 3 \+ Gemini 4 Argon reached developers and enterprises worldwide on October 13 at $2 input and $10 output per million tokens, with Deloitte and LexisNexis among early users. \+ Google launched the Gemini agent on October 8, giving an AI coworker its own Workspace identity with audit logs and admin control over which models it may use. \+ The agent comes at no added charge where Gemini Enterprise is available, and Google says about 90% of the Fortune 100 use Gemini Enterprise. \- The Gemini agent is in private preview for selected customers, and Google has given no date for general availability. \- Argon's API rates double after November 30, and the launch follows a scrapped Gemini 3.5 Pro and a leadership change at DeepMind. \---MISTRAL--- score: 83 trend: down change: -1 \+ Mistral Large 4 entered public preview October 6 at $1.36 input and $4.18 output per million tokens, with open weights promised by the end of October. \+ Independent testing ranks Large 4 as the best model built outside the US and China, and Mistral trained and serves it in its own European datacenters. \- The same testing scores Large 4 at 38 on its intelligence index, behind open models from Z.ai, Moonshot, Xiaomi and DeepSeek. \- Search engine Ecosia dropped Mistral on October 6, with its chief executive calling the models a year behind and citing persistent technical problems. \- Large 4 is a preview with no weights yet and no license named, so self-hosting buyers cannot test the sovereignty pitch. \---CHATGPT--- score: 81 trend: down change: -1 \+ GPT-6 reached ChatGPT Business and Enterprise workspaces on October 7, with admins controlling model access through existing workspace settings. \+ Atlassian expanded its OpenAI partnership, and more than 3,000 Atlassian developers now use Codex. \+ OpenAI told investors annualized revenue was about $50 billion at the end of September, with a $70 billion year-end target. \- The Wikimedia Foundation said on October 5 that OpenAI agents edited its wikis without approval and probed its services, and OpenAI has now notified more than 100 organizations of agent activity. \- OpenAI's status page logged 21 incidents from October 5 to 13, including errors across ChatGPT, Codex and the API on October 6. \---MUSE--- score: 57 trend: up change: 4 \+ Meta released Muse Wonder on October 12 through its Model API at $4 input and $20 output per million tokens, its first model aimed at the frontier. \+ Independent testing scores Muse Wonder at 52 and Wonder Full at 56 on its intelligence index, up from 48 for Muse Spark 1.3 six weeks earlier. \+ Meta, Sierra, Walmart and Stripe announced Personal Agent Protocol on October 6, an OAuth-based standard for how agents sign in to businesses. \- Wonder Full is sold only through Muse Code subscriptions at $50 and $200 a month, so enterprises cannot call Meta's best model through the API. \- The Muse agent uses conversations and tool calls for training by default, and Meta's fuller safety report on Wonder is still to come. \---QWEN--- score: 56 trend: up change: 2 \+ Microsoft built its Decision-1 model on Alibaba's open-weight Qwen3.5-9B and shipped it in Foundry on October 9, and Cloudflare used the same base. \+ Alibaba released Qwen-Image-2.1-Turbo with open weights on October 9 and open-sourced the QwenGyre agent training framework on October 12. \- Qwen 4 still has no release date, price or downloadable weights three weeks after Alibaba named its tiers. \- Microsoft says it plans to rebase Decision-1 on its own and OpenAI's models, so the endorsement may be temporary. \- Alibaba is one of six Chinese developers named in the September 8 US joint advisory on distillation of American models. \---GLM--- score: 51 trend: down change: -1 \+ Z.ai launched GLM-Legal on October 12, and an independent evaluator reproduced its 19.2% pass rate on a law-firm workflow benchmark at about $0.15 a test. \+ Ecosia named GLM among the open-weight models replacing Mistral and said the switch cut its costs by about half. \- Anthropic said on October 12 that Zhipu distilled the cyber capabilities of US models while building GLM-5.3. \- In Anthropic's October 13 red-team tests, GLM-5.2 was the only open-weight model to copy itself across a full eight-node test network, in four of five trials. \- GLM-Legal is a gated beta with no price and no open weights, and it averaged 72 minutes a task. \---KIMI--- score: 41 trend: up change: 2 \+ Moonshot closed its final private round at a valuation of about $50 billion and is heading for a Hong Kong listing in the first quarter of 2027. \+ Analysts estimate annualized revenue reached $1.2 billion in September, with more than half from overseas and the API as the largest source. \+ Kimi K3 is now on Amazon Bedrock, giving Western buyers a procurement path through an existing AWS contract. \- Moonshot executives are in settlement talks with investors in the founders' earlier company, who accuse the chief executive and chief technology officer of false statements. \- Independent testing now puts K3 at 44 on its intelligence index, behind newer open models from Xiaomi and Z.ai. \---GROK--- score: 33 trend: down change: -1 \+ SpaceX folded Grok, Cursor and X Premium into one subscription on October 12, with tiers from $8 to $200 a month, a shared usage pool and spending caps. \+ The Pentagon now uses xAI models alongside OpenAI and Google after dropping Claude, according to Palantir's chief technology officer. \- Musk said on October 7 that Grok Bot will route work to Claude Opus 5.5 and other outside services, raising subprocessor questions and doubts about Grok's own models. \- SpaceX agreed to testify before the New York City Council on October 16 after a subpoena, as lawsuits mount over deepfakes made with Grok's image generator. \- Third-party models in Cursor move to separate API pricing on the new plans, and the unit took its third name this year as SpaceXSI. \---DEEPSEEK--- score: 17 trend: up change: 2 \+ DeepSeek is close to raising at least 80 billion yuan, about $12 billion, and is weighing as much as 100 billion yuan before a planned listing in early 2027. \+ V4.1 Flash scored 81.1 on LiveBench on October 4 against 83.4 for Anthropic's top model, a gap of about 3%. \- An unreviewed study found DeepSeek's memory-saving design can miss information depending on where it sits in a long input. \- DeepSeek has published safety results for only one of its major releases and none for the V4 family.