---CHATGPT---
score: 88
trend: up
change: +3
+ Enterprise revenue passed the ChatGPT consumer business on Aug. 14, two quarters ahead of OpenAI's own parity forecast
+ Annualized run rate reached $40 billion, up 20% month over month in July, with business customer count up 32%
+ Advertising inside ChatGPT is approaching a $1 billion annual run rate
- Chief revenue officer Denise Dresser left after eight months, replaced by Wiz veteran Dali Rajic on Aug. 13
- Astra stays paused under tighter containment, and the four other services its agents reached are still unnamed
---CLAUDE---
score: 84
trend: down
change: -3
- The text watermark has no opt-out on any tier, including the platform API and Claude via AWS, Google Cloud and Microsoft Foundry
- Paying Claude Max subscribers at $100 a month are canceling over marks that survive proofreading and translation of their own writing
- Ramp's July sample puts Fable 5 model-attributed spending at roughly 75% of GPT-5.6 Sol
- StatusGator counts 164 outages since January, including a 7.5-hour disruption across four models on Aug. 5
+ First tracked vendor to actually implement EU AI Act Article 50 text marking, and still leads Ramp's paid-adoption sample
---GEMINI---
score: 82
trend: up
change: +3
+ Gemini 3.7 Flash shipped Aug. 13 at $0.75 and $3.75 per million tokens, half the prior model's launch price
+ Day-one availability across the Gemini Enterprise Agent Platform, AI Studio, Android Studio and Antigravity
+ Three-week cadence from 3.6 Flash, delivered through post-training rather than a fresh pretrain
- Gemini 3.5 Pro has now missed June, July and all of August, with reporting citing coding shortfalls and a weak data refresh
- Made by Google on Aug. 12 spent the quarter's biggest stage on consumer hardware
---MISTRAL---
score: 81
trend: up
change: +1
+ Regional Endpoints reached general availability, letting customers pin inference to Europe or the US
+ A Priority Tier in public preview adds Mistral's first committed uptime service level agreement
+ Compute coalition with Amadeus, ASML and CMA CGM targets 200 megawatts by end of 2027 and 1 gigawatt by 2030
- The gigawatt is pre-sold rather than built, and the roughly 3 billion euro round still has not closed
- Serving Z.ai's GLM-5.2 on its own platform complicates the sovereignty argument that wins Mistral its deals
---QWEN---
score: 50
trend: up
change: +4
+ Qwen3.8-2.4T weights landed on Hugging Face Aug. 13, keeping the open-weights commitment made at launch
+ Qwen3.8-27B followed Aug. 14 under Apache 2.0 and passed 3 million downloads
+ Apple switched Siri and Writing Tools on mainland China Macs to Qwen and is training its own China model with Alibaba
- The 27B ships with xhigh reasoning as the default, producing 20 to 90 minute completions on trivial prompts
- Beijing's weight-export consultation still points directly at the release Alibaba just made
---GLM---
score: 47
trend: up
change: +2
+ Z.ai held GLM-5.3 weights roughly two weeks after its own cyber tests came back higher than expected
+ Mistral now serves GLM-5.2 under European regional controls, the first Western compliance envelope around a Chinese open model
+ The launch ledger lists 2,436 vulnerabilities across 269 open-source projects, 1,097 rated critical or high
- The delay breaks the MIT-weights-within-days pattern that was Z.ai's clearest differentiator
- The 84.5% CyberGym result is from Z.ai's own harness, and the lead vanishes on exploitation at 54.4% against 78.0%
---MUSE---
score: 46
trend: up
change: +2
+ Muse Glimmer shipped Aug. 10 as a 30-billion-parameter Apache 2.0 model running local agents on one consumer GPU
+ The release reverses the closed-weight break from Llama and matches Meta's own lobbying position on open weights
+ Zuckerberg signaled Muse Spark 1.2 weights would follow
- The contributor tier still buys prompts and completions with a 12x input discount, unusable under any confidentiality duty
- The Aug. 5 Muse Spark 1.1 third-party breach stands on the record
---KIMI---
score: 39
trend: down
change: -1
+ The final pre-IPO round is closing near $30 billion, with the offshore structure being unwound for a Hong Kong listing
+ Annual recurring revenue moved from $200 million in April to more than $300 million in June, over 70% from API licensing
- Ten days after the sandbox-escape finding, Moonshot has still issued no statement to any outlet
- No new model, no license change, and no answer to US allegations on restricted chips and distillation
- Self-hosting the 2.8-trillion-parameter checkpoint still needs roughly 1.4 terabytes of memory
---GROK---
score: 32
trend: up
change: +5
+ Grok 4.6 shipped Aug. 12 at 61 on the Artificial Analysis Intelligence Index, tying GPT-5.6 Sol and two behind Opus 5
+ Day-one API access through xAI, Cursor, OpenRouter, Vercel and Cloudflare, ending the pattern of releases buyers could not reach
+ Pricing held at $2 and $6 per million tokens with a 500K context window
- The same-day model card is thin on autonomous behavior and safeguards, with no system card and no red-team report
- Requests above 200K tokens double to $4 and $12, and the Minnesota, UK and Memphis suits are unchanged
---DEEPSEEK---
score: 20
trend: down
change: -2
- API increases from 50% to more than 1,100% took effect Aug. 16, with V4-Flash output moving from a flat $0.28 to $1.32 at peak
- Peak windows run 01:00 to 04:00 and 06:00 to 10:00 UTC, so the European working morning pays the top rate
- V4-Pro output rises from $0.87 to $3.96 per million tokens at peak, ending the cost floor that was the entire buy case
+ The first outside funding round closed above $7 billion and IPO groundwork has begun
+ V4 Pro reached general availability Aug. 13 with 1.6 trillion parameters, 1 million token context and 384,000 output tokens

San Francisco

Editor-in-Chief and founder of Implicator.ai. Former ARD correspondent and senior broadcast journalist with 10+ years covering tech. Writes daily briefings on policy and market developments. Based in San Francisco. E-mail: editor@implicator.ai