Tityra
Daily brief ·

Google put Gemini 4 Argon in front of cyber defenders as the FTC opened an agent-safety probe

77.9% Gemini 4 Argon on DeepSWE v1.1, on Google's own numbers
$2 / $10 Gemini 4 Argon introductory price per million input / output tokens

TOP 3

  1. Google announced Gemini 4 Argon and is releasing it to trusted cyber defenders first rather than to the public.
  2. OpenAI and Synopsys signed a multi-year agreement to build GPT-Synopsys, a model specialised for chip design.
  3. The FTC has opened an investigation into OpenAI, Anthropic and METR over risks from agentic systems.

Models

  • Gemini 4 Argon: Google’s September 30 announcement describes a frontier model for coding, enterprise knowledge work and cyber defence, with a maximum output of 1 million tokens.
    • Access is phased, starting with trusted cyber defenders through Google’s Fairwind Program, and Google says it is engaged in the US government’s voluntary process for pre-release model access. Broader availability goes to paid API customers and AI Ultra subscribers, with no date given.
    • Launch pricing is $2 per million input tokens and $10 per million output tokens, with cached input at 95% off. The 77.9% on DeepSWE v1.1 is Google’s own figure.
  • GPT-Synopsys: Synopsys and OpenAI announced an expansive multi-year agreement to jointly develop a chip-design model, including a revenue-sharing arrangement and joint go-to-market.
    • OpenAI licenses Synopsys’ EDA tools to train the model to operate them, read their output and iterate on designs.
    • Synopsys states customer data is not used to train the model, is encrypted in transit and at rest, and is governed by configurable retention, audit and permission controls.
    • The release itself carries no performance figures. Speed and productivity numbers circulating in secondary coverage do not appear in the announcement.

Agentic AI & Agent Skills

  • Regulators reach agent behaviour: CNBC reports the FTC has opened an investigation into OpenAI, Anthropic and the evaluation group METR over risks from their AI products, and will issue civil investigative demands in the coming weeks.
    • This is reporting, not an FTC publication: the agency has not posted a statement, and the scope described comes from the outlet and its sources.

MCP & Plug-in

Nothing new

Tools & Open Source

  • CoreWeave Forge: CoreWeave announced Forge on September 30, a layer that joins running, observing, curating, improving and evaluating models and agents in one environment, bringing together Weights & Biases Models, post-training work from OpenPipe and the open-source marimo notebook project.
  • Sandboxes, generally available: CoreWeave Sandboxes gives each run an isolated environment for agent tool use, reinforcement learning or evaluation.

Community

Nothing new

Sources
  1. Gemini 4 Argon: our next era of frontier intelligence — Google
  2. OpenAI and Synopsys Announce GPT-Synopsys: Frontier Intelligence to Revolutionize Chip Design — Synopsys
  3. FTC is investigating OpenAI, Anthropic and other AI companies over product risks — CNBC
  4. CoreWeave Forge Connects the Full AI Development Loop — CoreWeave
  5. CoreWeave Sandboxes Launches to Accelerate Reinforcement Learning, Agent Tool Use and Model Evaluation — CoreWeave