AI NewsOctober 5, 2026

GPT-6 Astra, GPT-6.1 Sol, Gemini 4 Argon, and Claude Fable 5.1 all shipped within a month of each other, but their prices per task differ far more than their benchmark scores do.
Astra and Fable 5.1 share the same list price, $10 per million input tokens and $50 per million output tokens, while Sol and Argon cost one-fifth that. The gap that matters most for AI agents is cached-input pricing, the discounted rate for resending the same system prompt or tool instructions repeatedly: Astra charges $1.00 per million cached tokens, four times Fable 5.1's rate and ten times Sol's.
Argon remains the outlier structurally, with a 1 million-token output limit versus 128,000 for the other three, though it's only available through Google's limited Fairwind testing program for now, and its intro pricing is set to double later.
Key Capabilities:

• Price split: Astra and Fable 5.1 cost 5x more per token than Sol and Argon.
• Cache matters for agents: Astra's cached-input rate is 10x pricier than Sol's.
• Access gap: Argon is Fairwind-program-only; the other three are broadly available via API.

DeepSeek Harness now has official macOS and Windows apps, making its open-source "agent harness" (the runtime that lets an AI model read files, run commands, and follow a plan) easier to install than its original command-line version.
DeepSeek released the v0.2 preview on September 30, built on a plugin architecture where even the core agent loop can be swapped out. It bundles common work tools, a plugin manager, and a mode where the agent can write and install its own plugins based on a chat request. It also works with non-DeepSeek models through OpenAI-compatible connections, not just DeepSeek's own.
DeepSeek first open-sourced the tool in August under the MIT license, a permissive license allowing free commercial use. It's a developer preview, so DeepSeek warns that breaking changes will continue between versions.
Key Capabilities:

• New desktop apps: Official installers for macOS (Apple Silicon) and Windows; Linux users get it via npm.
• Self-extending: A "Creator" mode lets the agent build and install its own plugins mid-conversation.
• Model-agnostic: Works with other AI providers, not locked to DeepSeek's own models.

The newly announced Super Intelligence Force, led by Director of National Intelligence Jay Clayton, has 120 days to report on AI's risks and recommend the government's role.
Trump announced the Super Intelligence Force via Truth Social, naming Clayton alongside FTC Chair Andrew Ferguson, Pentagon CTO Emil Michael, and OPM Director Scott Kupor. The group reports directly to Trump and chief of staff Susie Wiles, and will coordinate federal engagement with AI companies, infrastructure providers, and the public.
The move follows a White House meeting where major AI companies signed a voluntary self-policing accord, and an executive order directing federal agencies to use the term "super intelligence" instead of "artificial intelligence." Clayton told the Wall Street Journal that "the risk of not being first is high."
Key Capabilities:

• Deadline: The task force has 120 days to deliver a report on AI risks and opportunities.
• Leadership: Led by DNI Jay Clayton, who keeps his existing intelligence role alongside this one.
• Scope: Will coordinate government engagement with AI companies, consumers, and infrastructure providers.

Product chief Tibo Sottiaux pledged that for the next 28 days, Codex and ChatGPT Work users get either a genuine improvement or a full usage reset, every single day.
OpenAI's Tibo Sottiaux said the team is focused only on four areas: simplifying features, improving usage efficiency, shipping new models, and building what he called "groundbreaking features." He said user feedback made clear people want the product to get simpler, not more complex.
No specific reset dates, eligible plans, or delivery details have been named, so whether or when your account gets a reset under this pledge isn't confirmed. OpenAI has issued usage resets before, including in August, June, and April of this year.
Key Capabilities:

• The pledge: One meaningful improvement or a full reset, every day, for 28 days (started October 4).
• Focus areas: Simplification, usage efficiency, new models, and major new features only.
• Not guaranteed: No specific reset dates or eligible plans have been confirmed for any account.