GPT-6.5 Leak? MiniMax M3.1 Flash, Gemini 4 Carbon and Kimi K3.1 Explained

Key takeaways

• The GPT-6.5 hint comes from a site once listed as an OpenAI parody, so treat it as a rumour
• GPT-6.1 Terra reportedly appeared in 18 Microsoft evaluation configs
• MiniMax M3.1 Flash is out, at a reported $0.15 / $0.60 per million input/output tokens, with open weights expected on 23 October
• Early Gemini 4 Carbon checkpoints did worse than the presenter expected in his coding tests
• Anthropic’s next model reportedly spots related bugs and asks before fixing them
• OpenAI says its models are better at modernising old code such as COBOL and Fortran
• Other items: Kimi K3.1 traces, GLM Omni research, Anthropic’s OSS Scanner, Nvidia speed gains and the Dream1 robot model

Another week, another wave of AI leaks. Some are worth your attention, and some are a website countdown and a lot of hope. This roundup has both, plus one release that could genuinely lower your AI bills.

Here is our plain-English summary of the WorldofAI video, what each story means for a small business, and a clear line between releases and rumours.

Is GPT-6.5 really coming?

There is no announcement, and the main clue is shaky. A website with a joke name, bought by OpenAI’s Tibo, reportedly hides a “6.5” reference in its code and a countdown to 1 November at 10am San Francisco time. That date is day 28 of OpenAI’s 28-day shipping pledge.

A firmer clue is that GPT-6.1 Terra reportedly replaced GPT-5.6 Terra in 18 evaluation configurations in a Microsoft pull request. The Information also reportedly expects a busy month of releases from OpenAI, Anthropic and Google.

What is MiniMax M3.1 Flash?

It is a fast, low-cost model from Chinese AI company MiniMax, now out of preview. MiniMax has confirmed that the “Space Money Alpha” model seen on leaderboards was an early version.

In the presenter’s tests, M3.1 Flash built a landing page for about 1 cent against $1.22 for Opus 5.5. He says the Opus result looked clearly better, and suggests using MiniMax to scaffold and a stronger model to refine. We turn that idea into a method in How to Draft With a Cheap AI Model and Refine With a Premium One.

How does Gemini 4 Carbon look so far?

Not great, in the presenter’s view. Business Insider reportedly said the Carbon checkpoints could match Claude Opus 5.5 at coding. After testing the Carbon B and C checkpoints, the presenter says they were faster than Gemini 4 Argon but no better, and sometimes worse, at coding. These are early checkpoints, so the final model could change.

What is reported about Anthropic’s next model?

A tester quoted by The Information says it understands what you are really trying to do. Ask it to fix one bug and it may spot that the bug belongs to a wider class of problems, then ask whether you want those fixed too. The presenter thinks this is Fable 5.5, which we covered in Claude Fable 5.5 Leaks: Is It Really “Baby AGI”? Plus GLM 5.5 and GPT-6 Astra Pro.

Anthropic has also launched OSS Scanner, which offers free regular security scans for important open-source projects. Each finding comes with a proof of concept and, where possible, a suggested fix.

What did OpenAI say about legacy code?

OpenAI says its latest models are much better at understanding and modernising old languages such as COBOL, Fortran and Visual Basic. Banks, government systems and plenty of ordinary businesses still run on this code, and the people who wrote it have often moved on.

This is one of the most practical stories in the roundup. If your business relies on an old system, our guide How to Plan a Legacy Code Modernisation With AI Coding Agents shows how to approach it safely.

Day 15 of OpenAI’s pledge was a livestream rather than a new release. It showed Codex cloud tasks that keep running after you close your laptop, and Codex CLI features for managing several agents.

What else was in the roundup?

We tested DeepSeek’s model in DeepSeek V4.1 Flash 1017 Tested: How a Cheap Open Model Stacks Up, and covered the weekend’s other news in Anthropic’s Public Benefit Fund: An AI UBI Plan? Plus OpenAI Aurora and OpenCode.

What is confirmed and what is not?

For a simple way to check leaks yourself, read How to Verify AI Model Leak Rumours Before You Switch Tools.

What should you do with this news?

1. Test MiniMax M3.1 Flash on a few of your real, low-risk tasks while free access lasts, and compare quality with your current model.
2. Ignore the GPT-6.5 countdown until OpenAI announces something.
3. List any old systems your business depends on, and consider whether AI could help document them.
4. Review your model mix with our AI model routing guide.

Frequently asked questions

The bottom line

The GPT-6.5 “leak” is mostly a countdown on a joke website, so do not plan around it. The story that matters most for small businesses is MiniMax M3.1 Flash, which could cut the cost of routine AI work if it holds up on your tasks. OpenAI’s push on legacy code is also worth watching if you depend on older systems. Start with our two practical guides.

Related posts

• Anthropic’s Public Benefit Fund: An AI UBI Plan? Plus OpenAI Aurora and OpenCode
• DeepSeek V4.1 Flash 1017 Tested: How a Cheap Open Model Stacks Up
• How to Verify AI Model Leak Rumours Before You Switch Tools

Garry Knight

I'm Garry Knight, the person behind Prodify Digital. I write about email list building, email marketing, SEO, AI search and the tools that connect them. My aim is to make online marketing easier to understand, so creators and small business owners can make informed decisions about building an audience and keeping people engaged. Here you'll find straightforward guides and product reviews that explain what something does, where it fits and which limitations matter. The focus is on clear explanations and useful next steps—not hype, shortcuts or promises of easy earnings.

See Full Bio