Frontier Model Releases and System Cards — 2026-09-02
Anthropic shipped the Fable/Mythos 5.1 model update alongside disclosures of security failures, while OpenAI prepared its "Astra" model with critical cyber capabilities. Google is reportedly finalizing a new Gemini model codenamed "Skimaki" that may surpass competitors in coding benchmarks, with a potential release as early as this week. Meanwhile, token spending indices have halved from their peak, intensifying price competition among major AI labs.
Frontier Model Releases and System Cards — 2026-09-02
Top developments
Anthropic Ships Fable/Mythos 5.1 and Discloses Security Failures
On September 1, 2026, Anthropic released updates to its Fable and Mythos 5.1 agent models. The release was accompanied by disclosures regarding security failures, marking a notable shift in how frontier labs communicate safety incidents alongside product launches. This update positions Anthropic's latest agent capabilities against recent moves by OpenAI and other competitors.
Google Prepares New Gemini Model Codenamed "Skimaki"
Google is reportedly close to launching a new Gemini AI model, internally codenamed "Skimaki," which could be released as early as the week of September 2, 2026. Sources indicate the new model may outperform rival models from OpenAI and Anthropic in coding benchmarks, specifically aiming to beat current leaders in software engineering tasks. This follows a period of rapid iteration for Google DeepMind, which recently updated its Gemini 3.7 Flash and other variants.
OpenAI Prepares "Astra" with Critical Cyber Capabilities
OpenAI is preparing a model named "Astra" that possesses critical cyber capabilities, according to reports from September 1, 2026. This development occurs as OpenAI and Anthropic executives have publicly argued for slowing down AI progress, even as they continue to develop more powerful systems. The Pentagon has also expanded military AI access during this period, highlighting the dual-use nature of these new capabilities.
Token Spending Index Halves, Driving Price Pressure
The LLM Token Spending Index, compiled by Silicon Data, has fallen by half from its peak, indicating a significant shift in market dynamics. This decline is exacerbating price competition among AI model providers, forcing labs to adjust pricing strategies to maintain volume. The pressure comes as new models like GPT-5.6 and Gemini 3.7 Flash continue to lower the cost-per-token baseline.
Local view
Japan: Business Insider Japan Highlights Agent Updates and Pricing
Business Insider Japan reported on September 1, 2026, that August saw significant updates to AI agents, including the introduction of "Grok Bot" and "Gemini 3.7 Flash." The article emphasized the trend of AI agents actively performing tasks rather than just generating text, and provided a comparative fee schedule for eight major generative AI services. The report notes that while performance improves, operational costs are being pushed down by competition.
China: Zhihu Tracks Model Updates and World Models
Zhihu, a leading Chinese Q&A platform, published an update on September 2, 2026, tracking global model releases. The post highlighted Anthropic's Claude Fable 5.1/Mythos 5.1 updates and noted the emergence of "World Labs" with a new world model named Atlas. The discussion reflects the growing importance of agent models, which now constitute a significant portion of the general-purpose category.
Context & numbers
Pricing and Performance Benchmarks
As of late August 2026, pricing for frontier models remains highly competitive. According to PromptZone, the cheapest frontier-adjacent tiers are:
- OpenAI: gpt-5.6-luna at $0.50/$3.00 per million tokens (input/output).
- Google: gemini-3.5-flash-lite at $0.30/$2.50 per million tokens.
- Anthropic: claude-haiku-4.5 at $1.00/$5.00 per million tokens.
Independent benchmark trackers like AI Release Tracker show GPT-5.4-Pro leading GPQA Diamond (graduate-level science reasoning) at 94.4%, while Claude Opus 4.7 leads SWE-Bench Verified (real-world software engineering) at 87.6%.
Industry Sentiment
Executives from both OpenAI and Anthropic spent August arguing that the industry needs to slow down, even as public sentiment sours on AI safety. Investor Gavin Baker noted on the All-In podcast that despite these calls for caution, development continues at pace.
On the radar
- TechCrunch Disrupt 2026: Anthropic and OpenAI are confirmed speakers at the AI Stage, scheduled for later in September. This event will likely feature further details on their respective safety and capability roadmaps.
- Google Gemini Launch: Keep an eye out for an official announcement regarding the "Skimaki" model or Gemini 3.8 Flash, which is rumored to be imminent.
This content was collected, curated, and summarized entirely by AI — including how and what to gather. It may contain inaccuracies. Crew does not guarantee the accuracy of any information presented here. Always verify facts on your own before acting on them. Crew assumes no legal liability for any consequences arising from reliance on this content.
Powered by
