Superintelligence Week: Gemini 4 Carbon Tops the Board, Terra Halves the Price, Pika Opens Its Video Model

Darlings, the machines had a busier Halloween week than most of us. Four releases between Oct. 26 and Oct. 30 change what a working creator can do on Monday morning: a new leader on the most-watched intelligence leaderboard, a fast model at half the price, an open video model with real sound mixing, and a privacy product aimed squarely at people who handle other people’s secrets. Here is what shipped, what it costs, and what we would actually do with it.

1. Gemini 4 Carbon takes No. 1, with an asterisk

What happened: Artificial Analysis reported on Oct. 28 that Gemini 4 Carbon is the new No. 1 on its Intelligence Index v5, scoring 59 and taking the top score in four of nine evaluations, including HLE-Diamond and Terminal-Bench-Science. On coding agents it lands second, three points behind Claude Code running Claude Opus 5.5. The asterisk: the firm notes Carbon is being rolled out to selected users and “is not yet publicly available,” with a 50 percent launch discount and no confirmed end date.

The security angle: The Verge reports Carbon is currently available only to a select group of cybersecurity defense companies, with a later expansion to partners in its Fairwind Program, a sign that frontier labs now treat vulnerability-finding as both a selling point and a hazard. What to do: if you are not on the access list, nothing yet. If you are, test it on your hardest research brief, not your easiest caption.

2. GPT-6.1 Terra: the workhorse gets cheaper

What happened: OpenAI introduced GPT-6.1 Terra, which it describes as a fast, efficient model for computer use, visual reasoning and “clearer, more natural writing.” Standard API pricing is one dollar per million input tokens and five dollars per million output tokens, which OpenAI says is 50 percent lower than GPT-5.6 Terra. Its own figures claim the error rate on its internal hallucination test fell from 53.6 percent to 16.7 percent at low reasoning effort. It is available to Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex.

RelatedHubble’s New Look at Messier 30, and NASA’s Faster Path Back to the Moon

What to do: move your repetitive jobs (invoice entry, form filling, shot logs, caption drafts) to the cheaper tier and keep the expensive models for the work that genuinely needs them. OpenAI itself says its larger models remain the better choice for the most complex coding.

3. Pika H3 Flex: open weights, five audio stems

What happened: On Oct. 29, Pika introduced Pika H3 Flex, a post-trained version of MiniMax H3 that accepts up to 12 references (images, video and audio) in one generation and, through the Pika API, exports five separate audio stems: Dialogue, Score, SFX, Foley and Ambience. Clips generate natively at 480p or 768p and can be upscaled to 1080p, 2K or 4K on Pika’s hosted service. Pika is releasing open weights and inference code, and says Flex arrives in its creation product on Nov. 2.

Why it matters to a nightlife desk: stems are the quiet revolution. A recap reel with the music too loud is the most common sin in party content; separate dialogue and score tracks let you fix the mix instead of reshooting. What to do: draft at 768p, upscale only the take you love, and keep your own photographs as the references so the look stays yours.

4. OpenAI Private Intelligence: privacy as a product

What happened: OpenAI’s new Private Intelligence page bundles its enterprise protections: Zero Data Retention for eligible workflows, Private Inference in API preview using confidential computing, and controls over what agents can access and when they need approval. “You own your business data, including your inputs and outputs,” the page says.

What to do: if you handle talent contracts, client lists or unreleased campaigns, ask your vendor in writing which of these protections your plan actually includes. Preview is not the same as available.

The habit everyone is copying

Per the stories trending on X, developers are writing standing instruction files (a CLAUDE.md in the project folder, for example) so their assistant remembers naming rules, folder rules and what never to touch. It is the most practical idea of the week, and it costs nothing: write one page of house rules, keep it next to your work, and update it when the assistant makes the same mistake twice. Eileen Otte Ford built Ford Models on exacting standards; your assistant deserves the same house rules.

Sources

1. Artificial Analysis: Gemini 4 Carbon is the new #1 on the Intelligence Index v5 (Oct. 28, 2026)
2. Gemini team blog: Introducing Gemini 4 Carbon
3. The Verge: Gemini 4 Carbon and the cybersecurity frontier
4. OpenAI: Introducing GPT-6.1 Terra
5. OpenAI developer docs: GPT-6.1 Terra
6. Pika: Introducing Pika H3 Flex (Oct. 29, 2026)
7. Business Wire: Pika launches Pika H3 Flex
8. OpenAI: Private Intelligence
9. X: Developers build standing instruction files for Claude

Nightlife Report is funded by readers. Back the reporting at patreon.com/aperlas.