AI Daily Brief · Oct 27, 2026 | OpenAI ships GPT-6.1 Terra; three science claims; NVIDIA/AMD buy Meta equity; Broadcom’s $63B Anthropic chip SPV

The past 24 hours swung from a major model launch to AI infrastructure finance. OpenAI introduced GPT-6.1 Terra—priced at about half the prior Terra tier, scoring 15 points higher than GPT-5.6 Terra on Agents’ Last Exam at roughly one-third the cost—and said its systems contributed to three scientific results spanning harmonic analysis, the Exponential Time Hypothesis, and quantum field theory. On the chip side, NVIDIA said it will invest $3 billion in Meta’s equity offering while expanding work on Muse, OpenShell, and Sentry; AMD separately pledged $1 billion and said Helios racks are already under test at Meta. Broadcom, Bank of America, Blackstone, Citi, and Morgan Stanley disclosed an about $63 billion SPV to buy Broadcom XPUs and lease them to Anthropic. Cursor released an Agent SDK and brought DeepSeek V4.1 Flash onto its platform; Claude for Nonprofits now counts more than 25,000 organizations.

Today’s highlights

• OpenAI shipped GPT-6.1 Terra: API pricing is $1 / $0.05 cached / $5 per million tokens; on Agents’ Last Exam at medium effort it beats GPT-5.6 Terra by 15 points at about one-third the cost; Terminal-Bench 4.0 at max effort reaches 50.3% (vs 21.5%).
• OpenAI announced three science claims (Kakeya / local smoothing, ETH, QCD/Yang–Mills); posts also say an internal Anthropic model independently refuted ETH—no official Anthropic confirmation [Unverified].
• NVIDIA will invest $3B in Meta and expand Muse / OpenShell / Sentry / Vera Rubin work; AMD will invest $1B and says Helios is installed at Meta.
• Broadcom and lenders structured ~$63B of financing via an SPV for Anthropic AI accelerators, with deliveries expected from 2027.
• Cursor launched an Agent SDK; DeepSeek V4.1 Flash is live in Cursor, with the company saying every effort tier sits on the CursorBench cost frontier.

1. OpenAI ships GPT-6.1 Terra: prices roughly halved, Agents’ Last Exam +15pp vs prior Terra at ~⅓ cost

Launch: OpenAI posted at 02:06 Beijing time on Oct 27 introducing GPT-6.1 Terra as a faster, more capable model for on-screen understanding and cross-app computer use, with more natural writing (25,913 likes, ~4.6M views). CEO Sam Altman followed up that “6.1 Terra is here,” highlighting computer-use performance (7,708 likes).

Pricing: Standard API rates are $1 per million input tokens, $0.05 per million cached input, and $5 per million output; OpenAI says that is 50% below standard GPT-5.6 Terra and GPT-6.1 Sol pricing. Artificial Analysis notes input/output prices are half or less of GPT-5.6 Terra’s $2/$12; cache writes are now $1.25 per million (halved) and cached input $0.05 (a quarter of the prior $0.20).

Benchmarks: On Agents’ Last Exam at medium reasoning effort, GPT-6.1 Terra scores 15 percentage points above GPT-5.6 Terra at about one-third the cost; it roughly matches GPT-6 Astra at about one-tenth the price per token. At max effort it is more than 3× as token-efficient as Claude Opus 5.5 and about one-fourth the price per token, for ~9× lower cost per task; it finishes tasks over 3× faster on average than Astra and GPT-6.1 Sol. On Terminal-Bench 4.0 at max effort it solves 50.3% of tasks versus GPT-5.6 Terra’s 21.5% (~2.3×). On an internal error-inducing prompt set, low-effort error rate fell from 53.6% to 16.7%, and max-effort error rate to 4.3%.

Third-party: Artificial Analysis calls it the first Terra-class model since GPT-5.6; at max effort its Intelligence Index is 6 points above GPT-5.6 Terra at $0.40 vs $1.40 per task; Coding Agent Index v2.0 scores 48 (Sol 54, Luna 31); on AA-Omniscience the guess rate when it does not know the answer fell from 88% to 38% [Self-reported by the leaderboard].

Availability: Rolling out to all Plus, Pro, Enterprise, and Edu users in ChatGPT Work and Codex; API name gpt-6.1-terra. GitHub says it is generally available in Copilot; OpenRouter listed it the same day.

Sources: @OpenAI 1 · @OpenAI 2 · @OpenAI 3 · @sama · @ArtificialAnlys 1 · @OpenAIDevs · @testingcatalog 1 · @github (379 likes) · @OpenRouter (163 likes) · OpenAI: Introducing GPT-6.1 Terra

2. OpenAI announces three scientific discoveries; an internal Anthropic model also said to refute ETH [Unverified]

Official claim: OpenAI research lead Greg Brockman said the company was announcing three major discoveries—in harmonic analysis (math), complexity theory (CS), and quantum field theory (physics)—linking to OpenAI’s science blog (13,385 likes, ~2.1M views). @danintheory posted in parallel (6,138 likes).

What the blog describes: Results related to the Kakeya set conjecture / local smoothing in all dimensions; a claim that the Exponential Time Hypothesis (ETH) is false along the Impagliazzo–Paturi–Williams line of work; and QCD / Yang–Mills–related results citing Gross and others. Brockman separately quipped that “p still probably doesn’t equal np.” These are company-and-collaborator accounts of research progress; independent peer review and publication are still underway [Company claim].

Anthropic parallel claim: @AndrewCurran_ said an internal Claude research model also independently refuted ETH on Oct 16 with ~14M output tokens and no human input, with Anthropic researchers then polishing the write-up (932 likes); @kimmonismus repeated the claim (677 likes). As of this brief there was no Anthropic official post or paper page confirming it [Unverified].

Sources: @gdb 1 · @danintheory · @gdb 2 (6,217 likes) · @kimmonismus 1 (1,199 likes) · @AndrewCurran_ · @kimmonismus 2 · OpenAI science blog

3. NVIDIA invests $3B in Meta to expand Muse; AMD invests $1B and says Helios is running at Meta

Background: Meta announced a plan on Monday (US time) to raise up to about $95 billion in equity (roughly $35B underwritten first, then up to ~$60B via an ATM) for AI infrastructure and superintelligence work. Coverage of that plan landed before this brief’s window and is background only here.

NVIDIA: The company account said NVIDIA and Meta will expand collaboration on Muse and the Muse family to improve inference and training; they are considering NVIDIA OpenShell for safe agent deployment and NVIDIA Sentry for monitoring agent behavior; Meta plans broader Vera Rubin and Vera CPU deployments with Spectrum-X Ethernet and BlueField-4 (1,713 likes). The press release states NVIDIA will invest $3 billion in Meta’s equity offering.

AMD: AMD’s Oct 27 release (9:20 a.m. ET, ~21:20 Beijing time) says Helios rack-scale systems are installed and being tested at Meta, EPYC processors power Muse agent environments, and AMD provides Day-0 support for Muse Spark Open and Muse Glimmer; AMD will also invest $1 billion in Meta’s equity offering. @testingcatalog summarized the NVIDIA–Meta points.

Combined: The two disclosed equity stakes total $4 billion ($3B + $1B). CNBC separately said chipmakers were interested in buying up to about $5 billion; formal company releases control.

Sources: @nvidianewsroom · @testingcatalog 2 (112 likes) · NVIDIA press release · AMD press release · CNBC (Meta $95B plan background) · CNBC (NVIDIA/AMD stakes)

4. Broadcom helps structure ~$63B financing so Anthropic can deploy AI chips

Official: A Broadcom investor release says Bank of America, Blackstone, Citi, and Morgan Stanley announced a $63 billion transaction via a new SPV that will buy Broadcom XPU AI accelerators and networking gear and lease them to Anthropic—part of Broadcom’s previously announced strategic compute deal, with deliveries expected starting in calendar 2027. First-lien financing is arranged by BofA, Citi, and Morgan Stanley; Blackstone alone provides the committed second-lien portion; Broadcom supplies residual-value support equal to the first-lien amount.

Media: @StockMKTNewz said Broadcom arranged a $63B deal for Anthropic chip financing with that SPV structure (349 likes). WSJ reporter Anissa Gardizy reported that CEO Hock Tan held weekend talks to lock in lenders, writing that lenders needed to commit about $40 billion and accepted higher rates (751 likes, ~243K views)—a different framing from the official “$63B structure,” so the official $63B figure is used here and the WSJ number is supplemental [Media report].

Sources: @StockMKTNewz · @anissagardizy8 · Broadcom / lenders release

5. Cursor ships Agent SDK; DeepSeek V4.1 Flash lands on Cursor’s cost frontier

Agent SDK: Cursor said the Agent SDK is available: define agents as code with prompts in Markdown and tools in TypeScript, deploy on Cursor-hosted or self-hosted infrastructure, with built-in evals (3,554 likes, ~335K views).

DeepSeek V4.1 Flash: The same account said DeepSeek V4.1 Flash is live in Cursor; “every effort level sits on the CursorBench cost frontier: from $0.60 to $0.98 per task, no model scores higher” (3,446 likes) [Company claim].

Sources: @cursor_ai 1 · @cursor_ai 2

6. Claude for Nonprofits expands: 25,000+ orgs, new fundraising and hiring skills

Official: Anthropic said more than 25,000 nonprofits now use Claude, mostly small local teams; it is expanding Claude for Nonprofits with a plugin built with about 30 of them, adding fundraising, marketing, and hiring skills grounded in each team’s context and AI policy (6,193 likes, ~621K views).

Sources: @claudeai · Expanding Claude for Nonprofits

In brief

• Artificial Analysis “Tomorrow” tease: @ArtificialAnlys posted only “Tomorrow.” (@ArtificialAnlys 2, 1,756 likes); @kimmonismus said even AA calling tomorrow a big day means something is coming (@kimmonismus 3, 1,919 likes). As of this brief there was no clear Intelligence Index v5 launch post; the same account also shipped Code2Video / AA-3D material the same day, so whether that was the tease remains unclear [Unverified].
• Ant Ling-3.1-flash-1019 open-sourced: @AntLingAGI said Ling-3.1-flash-1019 is open source at 44 on Artificial Analysis Intelligence Index v4.3.2 (up from 41), with a base release (@AntLingAGI, 245 likes) [Company claim / Self-reported by the leaderboard].
• Grok Bot: Elon Musk said Grok Bot is “like hiring an incredibly competent chief of staff” (@elonmusk, 23,623 likes, ~41.4M views) [Company claim]; @alx said he published bots to scrub spam-call lists and personal data (@alx, 3,659 likes).
• Jensen Huang at GTC Berlin: @JensenHuang posted that he was “the opening act for a family of ducks. Thank you, Berlin” (@JensenHuang, 5,509 likes, ~2.89M views).
• Fable 5.5 timing: @kimmonismus said Fable 5.5 is “probably not today but very very likely this week” (@kimmonismus 4, 638 likes); no Anthropic confirmation [Unverified].
• Vercel cpu-chat: Vercel CEO @rauchg said he reproduced OpenAI’s covert CPU-channel findings as “cpu-chat” (@rauchg, 2,385 likes).
• Dario on value-based pricing: @tmuxvim and @Indian_Bronson discussed Anthropic’s CEO on pricing AI by real-world value (@tmuxvim, 3,230 likes · @Indian_Bronson, 2,145 likes) [Secondhand account].
• Shorter thinking traces: @emollick said detailed thinking summaries for Claude and ChatGPT are shrinking, possibly anti-distillation (@emollick, 548 likes) [Personal observation].
• John Carmack on robot demos: @ID_AA_Carmack commented on flashy robot demos flooding timelines and urged against over-extrapolation (@ID_AA_Carmack, 5,642 likes) [Opinion].

---

About this column and how we verify

1. AI-assisted editing: AI Daily Brief is AI-assisted; this issue was produced with help from Grok Bot and covers Beijing time Oct 26, 2026, 22:05 through about Oct 27, 2026, 22:20.
2. Tweet-level checks: Every cited post was fetched to confirm it exists, the author matches, and the timestamp falls in this period; selected posts were also re-read via the X API where they came from X News clusters. Like and view counts are as of editing time.
3. X News summaries are leads only: X News blurbs are Grok-generated and used only to find stories; facts are checked against original posts, official pages, and mainstream coverage. When a summary conflicts with a primary source, the primary source wins.
4. Label key: [Unverified] = circulating without an official or other primary confirmation; [Company claim] = from a company release or official account, not independently reproduced here; [Self-reported by the leaderboard] = from a benchmark publisher’s own write-up; [Media report] = from press coverage that may differ in framing from the official release; [Secondhand account] = a retelling of someone else’s remarks; [Personal observation] = one person’s usage note; [Opinion] = judgment, not a factual claim.