# What actually mattered: Oct 1 to 9 > Source: > Published: 2026-10-12 16:00:00+00:00 Five things actually mattered in AI since October 1. Here they are, with what I think of each. **1. GPT-6 is live for everyone** - Anyone who has a paid ChatGPT plan now has access to GPT-6, and it’s starting to roll out to free users as well. - **My take:** It’s cool, but I’m no longer in love with GPT-6. Claude Opus 5.5 is far better for 99% of people, and way cheaper. **2. Claude Haiku 5.5 is here, and it’s cheap** - Anthropic released [its small, fast model](https://www.anthropic.com/claude-haiku-5-5) at about 75 percent cheaper pricing than the previous version of Haiku. - **My take:** This is a new class of very cheap, but super capable model, with very few compromises. I personally still wouldn’t daily-drive it, but for folks watching their AI spending, it’s an incredible option. I’d also recommend you try asking your Opus 5.5 agents to use Haiku as a sub-agent to speed up work and save on costs. **3. Anthropic opens its security tools** - Anthropic gave more defenders access to [its advanced security tools](https://www.anthropic.com/news/cyber-verification-program) . Its earlier partners found at least 129,000 verified software flaws between April and July. - **My take:** It’s very good that they’re expanding this program. My big worry with the original program was how few companies got access to it. We’ll see how it goes! - **Related:**[Axios reports](https://www.axios.com/2026/10/09/ai-companies-day-after-major-attack) that Anthropic, OpenAI and other labs are privately planning for “the day after” an AI-caused catastrophe, most likely a cyberattack on banking, the internet, or power and water. It’s... a little terrifying. **4. Google shows off Gemini 4 Argon** - Google announced [Gemini 4 Argon](https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-4-argon/) , but only a small group of security defenders can use it. There’s currently no release date for everyone else. - **My take:** Looks great on paper, but based on previous Gemini models, I’ll believe it when I see it. Gemini models tend to show great benchmarks, and look fantastic on paper, but fall apart in the real world. **5. OpenAI will watermark ChatGPT’s text in the EU** - OpenAI is adding [an invisible watermark](https://www.bleepingcomputer.com/news/artificial-intelligence/openai-is-adding-invisible-watermarks-to-chatgpt-and-codex-text-in-the-eu/) to ChatGPT’s text for users in the EU over the coming weeks, however, it’s not super strong, and very easy to bypass. — Matt What should I write about next? [Vote or suggest a topic →](https://somethingbig.ai/next)