- Published on
AI Weekly: OpenClaw Creator Joins OpenAI, Anthropic Raises $30B
- Authors
- Name
- Christopher Kvamme
- @MidnightBuild12
AI Weekly: OpenClaw Creator Joins OpenAI, Anthropic Raises $30B
TLDR
- OpenClaw creator Peter Steinberger is joining OpenAI. The tool will move to an open-source foundation (details TBD).
- Anthropic closed a 380 billion valuation.
- Claude Sonnet 4.6 launched February 17 and became the default for free and Pro Claude users.
- India's AI Impact Summit drew every major AI CEO. Altman confirmed 100 million weekly ChatGPT users in India.
- Alibaba released Qwen3.5 with agent and multimodal features. Agent tool compatibility is claimed but untested.
- OpenAI's Codex-Spark claims over 1,000 tokens per second for coding on Cerebras hardware (research preview).
- ByteDance is adding safeguards to Seedance 2.0 after pushback from six Hollywood studios and the MPA.
Table of Contents
- OpenClaw Creator Joins OpenAI
- Anthropic 380 Billion Valuation
- Claude Sonnet 4.6: Now the Default for All Users
- India AI Summit: 100 Million ChatGPT Users and Big Money
- Alibaba Releases Qwen3.5 with Agent and Multimodal Support
- OpenAI Codex-Spark: 1,000 Tokens Per Second on Cerebras
- ByteDance Adds Safeguards to Seedance 2.0 After Hollywood Pushback
- The Pattern This Week
- FAQ
- Conclusion
- Sources
Intro
Anthropic raised $30 billion. OpenClaw's creator joined OpenAI. Claude Sonnet 4.6 became the default model for every Claude user. All in one week.
Seven stories landed between February 11 and 17 that show where AI spending and talent are moving right now. Most of them point at the same thing: agents.
OpenClaw Creator Joins OpenAI
The most surprising story of the week did not come from a product launch.
On February 15, Sam Altman posted on X that Peter Steinberger is joining OpenAI. Steinberger created OpenClaw (previously Clawdbot and Moltbot). His role: "drive the next generation of personal agents." Altman called him "a genius with a lot of amazing ideas about the future of very smart agents."
What matters for users: OpenClaw will "live in a foundation as an open source project that OpenAI will continue to support." No terms were shared. CNBC confirmed the hire and noted his background as an Austrian developer known for PSPDFKit.
This fits a pattern. OpenAI reportedly paid over $6 billion for Jony Ive's AI devices startup in mid-2025. Hiring the creator of an open-source agent tool is a different bet, but the goal is the same: own the agents space.
Anthropic 380 Billion Valuation
On February 12, Anthropic [closed a 380 billion post-money valuation.
In the AI sector, only OpenAI's $40+ billion round led by SoftBank was larger. Coatue and Singapore's GIC led. D.E. Shaw, Dragoneer, Founders Fund, ICONIQ, and MGX also joined.
Part of the round includes earlier investments from Microsoft and Nvidia. For context, CNBC noted the valuation has surged from the company's earlier rounds. The scale of these raises reflects how much capital frontier AI labs need for compute infrastructure.
Alphabet reportedly disclosed plans to spend $175-185 billion on capex in 2026 during its Q4 earnings call. I keep coming back to this: the fundraising numbers only make sense if you account for how much compute costs. These are infrastructure bets, not software margins.
If you want to know which AI tools are worth paying for, our AI tools worth paying for in 2026 breakdown is a good place to start. The Anthropic Pro tier just got a big upgrade this week.
Claude Sonnet 4.6: Now the Default for All Users
Twelve days after Claude Opus 4.6 launched on February 5, Anthropic released Claude Sonnet 4.6 on February 17 and made it the default for all free and paid Pro users.
Anthropic's pitch: "Performance that would have previously required an Opus-class model is now available with Sonnet 4.6." The model is better at computer use, coding, design, and large data tasks.
The pace is the story. Two major models in 12 days. Anthropic is not doing big launch windows. It is shipping fast.
Every Claude user got a model swap without doing a thing. Anthropic's blog framed it as Opus-class results at Sonnet prices. That is Anthropic's claim, and independent benchmarks are still catching up. Whether the gap has truly closed will take weeks of real-world testing to confirm.
India AI Summit: 100 Million ChatGPT Users and Big Money
The India AI Impact Summit ran for five days starting February 16 at Bharat Mandapam.
Who showed up: Sam Altman, Dario Amodei, Sundar Pichai, Demis Hassabis, and Mukesh Ambani. PM Modi spoke alongside French President Macron, per TechCrunch.
The big number: Altman claimed India has 100 million weekly ChatGPT users, a self-reported figure from OpenAI. Altman said India is "second only to the United States" in usage. Altman also claimed India leads globally in student ChatGPT usage, per the same TechCrunch report.
The money was real. India put 1.2 billion to build domestic GPU infrastructure.
What stands out to me: India is not just using AI products. It is building the compute to run them domestically. That is a different posture than most countries take at AI summits.
Alibaba Releases Qwen3.5 with Agent and Multimodal Support
Alibaba released Qwen3.5 on February 16, the eve of Chinese New Year, amid what the company described as "intensifying competition" in China's AI space.
It ships in two forms: open-weight (the model weights are public, so anyone can download and run it on their own hardware) and hosted on Alibaba's servers. The model handles text, images, and video natively. CNBC reported the release pushed agent support and OpenAI-compatible APIs.
Here is what caught my eye: Alibaba pitched Qwen3.5 as compatible with open-source agent tools, the same week OpenClaw's creator joined OpenAI. The timing sends a message. But like the Sonnet 4.6 and Codex-Spark claims above, this is a vendor pitch. The compatibility claim is about API design, not a tested integration. No third-party test existed at the time of writing.
China's AI labs are moving as fast as US ones. The open-weight release keeps Qwen3.5 available to developers everywhere. That is a real edge over closed-model providers.
OpenAI Codex-Spark: 1,000 Tokens Per Second on Cerebras
OpenAI released GPT-5.3-Codex-Spark as a research preview for ChatGPT Pro users this week. Cerebras confirmed it runs on their Wafer Scale Engine 3 chips. TechCrunch reported OpenAI positioned it as a "daily productivity driver" for rapid prototyping.
The pitch is speed, not smarts. Cerebras claims Codex-Spark delivers over 1,000 tokens per second on their hardware. If accurate, that is fast enough for real-time code editing. Independent speed tests have not been published yet.
It is text only at launch, built specifically for code generation.
What I find interesting here: most new OpenAI models are bigger. Codex-Spark is faster instead. It is narrow, built for coding, and runs on dedicated silicon. For comparison, Groq has been hitting similar token-per-second numbers on smaller open models using its LPU hardware. But Codex-Spark is a GPT-5 class model at those speeds, not a 7B parameter model. If the speed claims hold up in independent testing, this is a meaningful step for real-time code editing workflows where latency matters more than reasoning depth.
The coding AI market is getting crowded fast. Our best AI code assistants for 2026 guide covers the current field. For the ongoing Cursor vs. Copilot debate, our Cursor vs. GitHub Copilot comparison has the latest take.
ByteDance Adds Safeguards to Seedance 2.0 After Hollywood Pushback
ByteDance announced it will add safeguards to Seedance 2.0 after backlash from Disney, Netflix, Paramount Skydance, Sony, Universal, Warner Bros. Discovery, and the Motion Picture Association.
The coalition is notable. Six studios plus an industry trade group pushing back at once. The concern is copyright in AI-generated video. ByteDance gave no timeline or technical details on what the safeguards will look like, per CNBC. That is the key gap. "Safeguards" could mean anything from content ID filtering to prompt-level blocking to licensing deals.
The pattern is familiar. Stability AI added content filters after Getty Images sued in early 2023. Midjourney blocked prompts targeting specific artists. Now ByteDance is making the same move for video: compliance over confrontation. The difference is scale. Video generation costs more compute and creates more obvious derivative works than still images. The studios know that, which is why they coordinated rather than filing individually.
TikTok needs those same content owners on its side. ByteDance cannot afford a licensing war with Hollywood while also navigating its ongoing US regulatory situation.
Our AI image generators breakdown covers where the video and image generation market sits now. Our sister site EmergentWeirdness goes deeper on the tension between AI tools and creative industries.
The Pattern This Week
Most of these stories share a thread: agents.
Altman hired Steinberger explicitly for agent work. Alibaba put agent capabilities in the headline of a major release. Anthropic led the Sonnet 4.6 announcement with "computer use." Codex-Spark is not a chatbot. OpenAI built it to execute tasks in real time.
The India Summit tells a slightly different story. It is about market scale and infrastructure investment. But the GPU buildout India is funding will power the same agent workloads everyone else is chasing.
The ByteDance story is the exception. It is about copyright, not agents. But it still fits a broader theme: AI tools are now good enough to threaten established industries. That is what triggers both the agent gold rush and the IP backlash.
The OpenClaw hire, the Qwen3.5 agent APIs, and Codex-Spark's real-time coding model all signal that the agent bet is becoming concrete. These are hiring decisions, product releases, and research previews, not finished products yet. But the direction is unmistakable.
The direction is clear. The timelines and details are not. That is what makes next week interesting.
FAQ
Is OpenClaw still safe to use now that its creator is going to OpenAI?
Altman said OpenClaw will "live in a foundation as an open source project." The code is not going behind a wall. But the foundation structure, licensing terms, and contributor agreements have not been announced yet. Until those details are public, the "open source" promise is a statement of intent, not a legal guarantee. Watch the foundation announcement.
What does the Anthropic $30 billion round mean for Claude pricing?
Anthropic's announcement said nothing about pricing. The round is about compute and research. No free tier or Pro tier changes were disclosed. The same week, Anthropic made Sonnet 4.6 the free default, which suggests they are competing on value. But funding rounds and pricing are separate decisions. Do not assume one drives the other.
How does Codex-Spark compare to existing fast-inference options?
Cerebras claims 1,000+ tokens per second. Groq hits similar speeds on smaller open models using its LPU hardware, but Codex-Spark is a GPT-5 class model, not a 7B parameter one. It is text-only and tuned specifically for coding. For real-time code generation, the speed matters. For general-purpose AI work, the GPT-5.2 family is still the better pick.
Conclusion
Two stories I am tracking into next week: the OpenClaw foundation announcement (governance structure and licensing terms are still unknown) and what Anthropic does first with $30 billion in fresh capital. The India AI Summit continues through February 20, so more announcements may surface by Wednesday. I will cover whatever develops.
Sources
CNBC, February 15, 2026: OpenClaw creator Peter Steinberger joining OpenAI, Altman says
Sam Altman on X, February 15, 2026: Steinberger joining OpenAI announcement
Anthropic (official), February 12, 2026: Anthropic raises 380 billion valuation
CNBC, February 12, 2026: Anthropic closes $30 billion funding round
CNBC, February 17, 2026: Anthropic releases Claude Sonnet 4.6
TechCrunch, February 16, 2026: All the important news from the ongoing India AI Impact Summit
TechCrunch, February 15, 2026: India has 100M weekly active ChatGPT users
TechCrunch, February 14, 2026: India doubles down on state-backed venture capital
TechCrunch, February 15, 2026: Blackstone backs Neysa in up to $1.2B financing
CNBC, February 17, 2026: Alibaba unveils Qwen3.5 as China's chatbot race shifts to AI agents
Alibaba Qwen Blog: Qwen3.5 Release
Cerebras Blog, February 2026: Introducing OpenAI GPT-5.3-Codex-Spark
TechCrunch, February 12, 2026: A new version of OpenAI's Codex is powered by a new dedicated chip
CNBC, February 16, 2026: ByteDance says it will add safeguards to Seedance 2.0
CNBC, February 5, 2026: Anthropic launches Claude Opus 4.6
CNBC, February 4, 2026: Alphabet Q4 2025 earnings