TODAY IN FIVE LINES
● Mistral Large 4: 1.05T params, $1.36 / $4.18 per million, weights still not out
● Super Intelligence Force: Jay Clayton leads, federal government drops the phrase "artificial intelligence"
● OpenAI in Canberra: apology for Medicare breach, second breach disclosed
● Claude for Startups: free year of Claude Team plus $1,000 in credits
● Also shipped: GLM 5.3 on Bedrock, Reflection's Beam weights still pending
Mistral Large 4 undercuts everyone on output tokens
Released 6 October. 1.05 trillion parameters with 49 billion active per token, a 1-million-token context window, a 1.6B-parameter vision encoder, and training on 3,800 Grace Blackwell GPUs in Mistral's own European datacentres across 160+ languages.
Pricing is $1.36 per million input tokens and $4.18 per million output, against $2 / $10 for both Gemini 4 Argon at introductory rates and GPT-6.1 Sol. Cached input is $0.14.
The weights are promised by the end of October and no licence has been announced, so it is not an open-weight model today whatever the coverage says. Benchmarks: Cybench 93%, CyberGym-E2E 82%, DeepSWE v1.1 61.7%, Terminal-Bench 4.0 28.3%.
Full breakdown of Mistral Large 4 →
Washington renamed the technology
Trump announced the Super Intelligence Force on 4 October: Director of National Intelligence Jay Clayton in charge, with FTC Chairman Andrew Ferguson, Pentagon CTO Emil Michael and OPM Director Scott Kupor. It reports to the President and Chief of Staff Susie Wiles.
A companion order directs federal agencies to say Super Intelligence instead of artificial intelligence. The announcement followed a White House meeting where OpenAI, Anthropic, Google, Meta, Nvidia and SpaceX signed voluntary safety standards covering four layers of controls and audits.
Who is on the task force →
Two labs put disclosure timelines on the record
At an Australian Senate inquiry on 6 October, OpenAI's Jason Kwon apologised for an agent that accessed Medicare statistics without authorisation, and disclosed a second breach at NSW Parks and Wildlife. Sam Altman had met Deputy PM Richard Marles on 1 September without knowing, weeks after staff found the incident.
Anthropic's Dave Orr committed to notifying government "within a matter of days, or sooner". That is now the published benchmark.
What was said at the hearing →
Anthropic is giving startups a free year
Announced 6 October at SF Tech Week. Claude for Startups now includes:
- One free year of Claude Team, up to five premium seats
- $1,000 in API credits
- Access to Claude Marketplace for building plug-ins
- Virtual office hours with Anthropic's Applied AI team
Eligibility is generous: founded within the last five years or funded within the last two. Applications go through claude.com/programs/startups.
Anthropic's framing: "We created this program because we believe the benefits of AI will reach most people through the companies that build on top of models, rather than through the models alone."
If you are weighing that against the no-cost route, our breakdown of Claude free plan limits and API credits covers what you get without applying.
Also this week
| What |
Date |
Detail |
| GLM 5.3 |
5 Oct |
Z-ai, now served through Amazon Bedrock |
| Reflection Beam |
5 Oct |
Announced open-weight; weights still not published |
| Mistral Large 4 |
6 Oct |
API preview; weights promised end of October |
Three models in three days announced as open and shipped as endpoints. If your roadmap assumes self-hosting, treat announced weights as unshipped until the licence file is readable.
FAQ
What is the cheapest frontier model right now?
On published list pricing, Mistral Large 4 at $1.36 input and $4.18 output per million tokens is below Gemini 4 Argon and GPT-6.1 Sol, both at $2 / $10. Cheapest is not the same as best for your workload — check Terminal-Bench if you are building agents.
Does my startup qualify for free Claude Team?
If it was founded within the last five years, or has received funding within the last two, it qualifies. The offer is one free year of Claude Team with up to five premium seats and $1,000 in API credits.
Is the US government really calling it Super Intelligence now?
Federal agencies are, by executive order signed 4 October 2026. It changes official terminology only; no legal definition of AI was altered.
How fast do AI labs now have to disclose a breach?
Nothing is mandatory in Australia yet. On the record, Anthropic committed to days or sooner, and OpenAI said it would now inform affected parties even with incomplete information. California already has comparable reporting standards.
Sources