Skip to content
FAQ
AI & ML Analysis

AI Week, Feb 2–8, 2026: SpaceX Absorbs xAI, and Anthropic and OpenAI Ship Rival Models on the Same Thursday

SpaceX acquired xAI with Musk pitching space-based AI data centers, Anthropic released Claude Opus 4.6 and OpenAI released GPT-5.3-Codex on February 5, and Benchmark put at least $225 million into Cerebras ahead of a planned Q2 2026 IPO.

6 min read

Written on November 15, 2026, covering news from Feb 2–8, 2026.

The week opened with Elon Musk folding his AI company into his rocket company, on the argument that AI’s power appetite eventually has to be met in orbit. Two days later, Anthropic and OpenAI each released a new model aimed at longer, more autonomous work. By the weekend, one of Cerebras’ earliest backers had set up dedicated vehicles to buy more of the chipmaker before it goes public.

Tuesday, February 3

SpaceX has acquired xAI, Musk announced in a statement posted on the SpaceX website on Tuesday. Yahoo Finance, in a report dated February 2, said Bloomberg had first reported the merger earlier on that Monday.

The deal puts xAI, best known for the Grok chatbot, alongside SpaceX’s Falcon and Starship rocket programmes. Al Jazeera noted that xAI itself acquired X, the former Twitter, last year, so the combination also pulls a social platform into SpaceX. Musk called the merged company “the most ambitious, vertically-integrated innovation engine on (and off) Earth”.

The power argument

Musk framed the merger around energy. He said AI needs “immense amounts of power and cooling” that cannot be supplied on Earth without “imposing hardship on communities and the environment”. In the version of the statement quoted by Yahoo Finance, he wrote that “Global electricity demand for AI simply cannot be met with terrestrial solutions, even in the near term”. He also predicted that within the next “2 to 3 years, the lowest cost way to generate AI compute will be in space”.

The hardware plan behind that claim is large. Musk said his long-term goal for SpaceX is to launch a million satellites, and that Starship aims to eventually fly once an hour carrying a 200-tonne payload. He added that Starlink’s next-generation V3 satellites will each add “more than 20 times the capacity to the constellation as the current Falcon launches of the V2 Starlink satellites”.

Musk is not alone in this idea. Al Jazeera reported that Jeff Bezos’s Blue Origin and Google’s Project Suncatcher each have orbital data-centre efforts that would run on solar power.

Government ties and the money

Both SpaceX and xAI hold major contracts with US agencies including NASA and the Department of Defense, and SpaceX’s Starshield unit works with military and intelligence bodies.

On finances, Yahoo Finance cited Bloomberg News as reporting that SpaceX is planning an IPO that could raise as much as $50 billion at a valuation of about $1.5 trillion. The outlet said it is unclear how those figures will shift with xAI now inside SpaceX. Yahoo Finance also reported that, in the week before its story, Tesla confirmed a $2 billion investment in xAI preferred stock.

Thursday, February 5

Anthropic: Claude Opus 4.6

Anthropic released Claude Opus 4.6, which CNBC described as the company’s first major model launch of the year, arriving months after Opus 4.5, Sonnet 4.5 and Haiku 4.5 late last year. It is the first Opus-class model with a 1M-token context window, offered in beta.

Anthropic’s claims centre on coding and long tasks: more careful planning, better code review and debugging, and steadier work in large codebases. The company said it posts the top score on Terminal-Bench 2.0 and leads other frontier models on Humanity’s Last Exam. On GDPval-AA, a test of economically valuable knowledge work, Anthropic said Opus 4.6 beats OpenAI’s GPT-5.2 by around 144 Elo points and Opus 4.5 by 190. On the 8-needle 1M variant of the MRCR v2 long-context test, it scored 76%, against 18.5% for Sonnet 4.5. CNBC added that the model now holds the top spot on the Finance Agent benchmark.

Alongside the model came several developer controls:

  • Adaptive thinking, letting the model decide when deeper reasoning helps.
  • 4 effort levels: low, medium, high (the default) and max.
  • Context compaction in beta, which summarizes older context as a conversation nears a set threshold.
  • Output of up to 128k tokens.
  • US-only inference at 1.1× token pricing.

Base pricing stays at $5/$25 per million tokens, with premium rates of $10/$37.50 per million input/output tokens for prompts over 200k tokens, available only on the Claude Platform. Claude Code gained agent teams as a research preview, and Claude in PowerPoint entered research preview for Max, Team and Enterprise plans.

On safety, Anthropic said the model is as well-aligned as Opus 4.5 and has the lowest over-refusal rate of any recent Claude model. Because of its stronger cybersecurity abilities, the company built 6 new cybersecurity probes to track misuse, and said it may introduce real-time intervention to block abuse in the near future.

The business backdrop matters here. CEO Dario Amodei told CNBC the previous month that enterprise customers make up roughly 80% of Anthropic’s business. CNBC also reported that Claude Code and Claude Cowork have unsettled software investors, and that the WisdomTree Cloud Computing Fund has fallen more than 20% since the start of the year. Scott White, Anthropic’s head of product for enterprise, told CNBC: “I think that we are now transitioning almost into vibe working”.

OpenAI: GPT-5.3-Codex

On the same day, OpenAI launched GPT-5.3-Codex. Digital Applied reported that it merges GPT-5.2-Codex’s coding performance with GPT-5.2’s reasoning and professional knowledge in one model that runs 25% faster. According to the same outlet, OpenAI describes it as the first model that was instrumental in creating itself, with early versions used to debug training and manage deployment.

OpenAI’s comparison against GPT-5.2-Codex, as reported by Digital Applied, shows the biggest gains on terminal and computer-use work:

  • Terminal-Bench 2.0: 77.3% vs 64.0%.
  • OSWorld-Verified: 64.7% vs 38.2%.
  • Cybersecurity CTF: 77.6% vs 67.4%.
  • SWE-Lancer IC Diamond: 81.4% vs 76.0%.
  • SWE-Bench Pro Public: 56.8% vs 56.4%.

The model is available to paid ChatGPT plans across the Codex app, CLI, IDE extension and web, with no free plan and no open weights. API access was listed as coming in the following weeks, with no exact date given at launch.

Digital Applied reported that GPT-5.3-Codex is the first model OpenAI rates as High capability for cybersecurity under its Preparedness Framework, with advanced cyber use gated through vetted trusted-access workflows. OpenAI paired the release with a Trusted Access for Cyber pilot, a wider private beta of its Aardvark security agent, and a $10M commitment in API credits for defending open-source software and critical infrastructure.

Also this week

Reported on February 7: Benchmark Capital put at least $225 million into Cerebras Systems through two vehicles, both named ‘Benchmark Infrastructure’, according to regulatory filings and a person familiar with the deal. The money was part of a $1 billion round, led by Tiger Global, that values Cerebras at $23 billion, nearly triple the $8.1 billion valuation from six months earlier. Benchmark keeps its funds under $450 million, which is why the separate vehicles were set up for this investment. Benchmark declined to comment.

Benchmark led Cerebras’ $27 million Series A in 2016. The chipmaker’s Wafer Scale Engine measures about 8.5 inches per side, holds 4 trillion transistors and 900,000 cores, and the company claims inference runs more than 20 times faster than competing systems. The previous month, Cerebras signed a multi-year deal worth more than $10 billion to supply 750 megawatts of compute to OpenAI through 2028; OpenAI CEO Sam Altman is also a Cerebras investor.

The road to listing has been rough. G42, a UAE-based AI firm, made up 87% of Cerebras’ revenue in the first half of 2024, and its past ties to Chinese tech companies triggered a CFIUS review that led Cerebras to withdraw an earlier filing in early 2025. G42 was off the investor list by late last year, and Reuters reports Cerebras is preparing to go public in the second quarter of 2026.

What to Watch

  • SpaceX’s IPO math. Yahoo Finance said it is unclear how Bloomberg’s figures of up to $50 billion raised at about $1.5 trillion will shift now that xAI sits inside SpaceX. I’d watch for revised numbers before treating them as settled.
  • GPT-5.3-Codex on the API. OpenAI gave no date or pricing for API access at launch. Until that lands, my read is that direct cost comparisons with Opus 4.6’s $5/$25 pricing are not possible.
  • Terminal-Bench 2.0 claims. Anthropic says Opus 4.6 has the top Terminal-Bench 2.0 score, while OpenAI reports 77.3% for GPT-5.3-Codex. The sources here do not put the two on one chart, so independent side-by-side results are the thing to look for.
  • Cerebras in Q2 2026. Reuters says the listing is planned for the second quarter. The open question, in my view, is how investors weigh the OpenAI contract against the G42 concentration history.

This article was written by AI from the sources listed on this page. Figures, names and quotes were checked against those sources by software before publishing.

SpaceX xAI Anthropic Claude Opus 4.6 OpenAI GPT-5.3-Codex Cerebras Benchmark Capital
Share