On February 5, 2026, Anthropic launched Claude Opus 4.6. The impact was immediate: $285 billion in market capitalisation evaporated from the software sector within hours. A week later, Anthropic raised $30 billion at a $380 billion valuation. Two opposite reactions to the same event. Both are rational.
What Anthropic put on the table
Opus 4.6 is not a point release. Three innovations place it in a category of its own.
First, Agent Teams: multiple AI agents divide up a complex project and work in parallel. Scott White, Anthropic’s Head of Product, compares the process to a human team coordinating in real time. One agent handles the frontend while another refactors the backend and a third writes documentation.
No external orchestration required.
Second, a one-million-token context window in beta.
Three thousand pages in a single pass. An entire codebase, a full audit file, a set of contracts — everything fits in one request.
Third, the model excels where economic value is highest. On GDPval-AA (finance, legal, analysis benchmark), Opus 4.6 outperforms GPT-5.2 by 144 Elo points. On Terminal-Bench 2.0 for agentic coding: best score in the industry.
In cybersecurity, the result is more striking still: the model autonomously identified over 500 zero-day vulnerabilities in open-source code, a rate exceeding that of most human audit teams across comparable scopes.
The progress that matters here is not the prettier sentence. It is the more complete task.
Why the market sold off
On launch day, the iShares Expanded Tech-Software ETF (IGV) dropped nearly 5%.
Since the start of 2026, the decline exceeds 22%, according to CNBC and Bloomberg. The panic goes well beyond Opus 4.6: investors are reassessing the viability of the traditional SaaS model when AI agents accomplish tasks that previously required per-seat software licences.
Salesforce: down 28%. Atlassian: down 35%. Thomson Reuters: down 16% in a single day.
The reasoning is straightforward. If ten AI agents replace the work of a hundred salespeople, a company no longer needs a hundred Salesforce licences. The per-seat revenue model — the backbone of SaaS for twenty years — is being challenged.
« Get me out. » Two words from Jeffrey Favuzza, trader at Jefferies.
Between January and February 2026, approximately $2 trillion in market capitalisation evaporated from the software sector, according to multiple converging estimates. But not everyone is selling.
Goldman Sachs considers the correction « too broad ». JPMorgan identifies solid fundamentals beneath the panic. Jensen Huang, NVIDIA’s CEO, called the idea that AI would replace the software industry « the most illogical thing in the world »: AI needs software to run.
The market is not saying « everything will vanish ».
It is saying: what used to be a product advantage is becoming an execution advantage.
$30 billion raised: Anthropic joins the global top three
A week after launch, Anthropic closed a $30 billion Series G round. The second-largest private funding in tech history, behind OpenAI’s $40 billion.
GIC (Singapore’s sovereign wealth fund) and Coatue led the round, alongside Microsoft, NVIDIA, Founders Fund and 36 other investors.
Anthropic’s annualised revenue now stands at $14 billion. Growth rate: 10x per year over the past three years.
Claude Code alone generates $2.5 billion in run-rate — a figure that has doubled since January 2026. Eight Fortune 10 companies are clients. Over 500 companies spend more than $1 million a year (two years ago, there were about a dozen; the trajectory is vertiginous).
Fortune places Anthropic alongside OpenAI and SpaceX. The question of an IPO is now being raised publicly.
Professionals reacted differently: « Right, this is becoming usable »
Practitioners do not have a valuation-multiple reflex. They have a workflow reflex.
Scott White uses a term that captures the shift: « vibe working ». After « vibe coding » (building software by intent rather than syntax), Opus 4.6 extends the logic to all knowledge work. Finance, law, research, analysis.
You define what you want. The agents execute it.
The clearest example is legal. At Dentons, the world’s largest law firm, CTO Matej Jambrich confirms that Opus 4.6’s reasoning capacity reduces rework and improves document consistency. Harvey reports a 90.2% score on BigLaw Bench, with 40% of tasks rated perfect.
That kind of signal speaks to quality and stability, not hype.
On Hacker News, one developer describes the new reality: spending more time in review mode than in coding mode. Agents implement while he validates the previous output.
At v0 (Vercel), a single test says more than any benchmark: Opus 4.6 produced a functioning physics engine in one pass.
The consensus is not universal. Several developers note that creative writing quality has slightly declined compared to Opus 4.5. A trade-off that engineering teams consider acceptable, but one that creative teams are watching closely.
Sonnet 4.6: democratisation, two weeks later
On February 17, Anthropic followed up with Claude Sonnet 4.6, the new default model for all users — free and Pro alike.
The key point: Sonnet 4.6 delivers performance comparable to Opus 4.5 at a fraction of the cost. Three dollars per million input tokens versus five for Opus.
The flagship performance of three months ago is now the free standard.
For European companies, the implication is concrete: the capability level that required an Opus budget is now available in Sonnet. The gap between tiers is being deliberately compressed, and that compression changes the equation for every marketing, editorial and analytical use case.
What it means for marketing and growth teams
If raw capability advances at this pace, value shifts to what the model does not « own »: proprietary data and business context, integrations and compliance, orchestration.
Three practical implications.
Agentic AI moves from demo to daily use. Agent Teams enable the parallelisation of complex workflows: research, structuring, production, validation. For a marketing team, sequential tasks become simultaneous ones. But the tool alone is no longer enough.
The SaaS business model is changing. If your stack includes fifteen per-seat tools, the consolidation question is on the table now. Agents do not yet replace all software, but the average investor has already started pricing in this transition. Framing makes the difference.
The quality of framing becomes the competitive advantage. The more powerful the model, the more the quality of the brief, the standards and human validation determine the output. A high-performing agent without editorial guardrails produces fast and badly. A high-performing agent inside a structured system produces fast and well — and that is exactly what we work on at Fast Growth Advisors with NOMO IA.
The more AI can do, the more companies pay for control.
Summary
Opus 4.6 generated more movement in two weeks than most tech launches in all of 2025. A historic market correction, a record funding round, a vocabulary shift across the industry.
For marketing and growth teams, one clear signal.
The window to structure your AI workflows is closing fast. Companies that have editorial frameworks and clear standards will absorb this acceleration. The rest will be subjected to it.
Sources
- CNBC — Anthropic Claude Opus 4.6 « Vibe Working » (Feb. 5, 2026)
- CNBC — AI Tools, SaaS Software Stocks Selloff (Feb. 6, 2026)
- CNBC — Anthropic $30B Round at $380B Valuation (Feb. 12, 2026)
- CNBC — Fears of AI Disruption: Software Options Trade (Feb. 13, 2026)
- CNBC — Claude Sonnet 4.6 Default Free Pro (Feb. 17, 2026)
- Bloomberg — What’s Behind the « SaaSpocalypse » (Feb. 4, 2026)
- Bloomberg — Anthropic $30B at $380B Value (Feb. 12, 2026)
- VentureBeat — Opus 4.6: 1M Token Context + Agent Teams (Feb. 5, 2026)
- TechCrunch — Anthropic Releases Opus 4.6 (Feb. 5, 2026)
- TechCrunch — Anthropic $30B Series G (Feb. 12, 2026)
- Fortune — Anthropic $30B, AI Arms Race (Feb. 13, 2026)
- Fortune — $380B Valuation, IPO Candidates (Feb. 13, 2026)
- Crunchbase — Second Largest Deal All Time (Feb. 12, 2026)
- Microsoft Azure — Opus 4.6 in Foundry (Feb. 2026)
- Anthropic — Claude Opus 4.6 Official (Feb. 5, 2026)
- Anthropic — Series G Announcement (Feb. 12, 2026)
FAQ
What exactly is Opus 4.6?
Claude Opus 4.6 is Anthropic’s most advanced model, launched on February 5, 2026. It introduces Agent Teams (AI agents working in parallel), a one-million-token context window, and state-of-the-art performance on coding, finance and legal benchmarks.
Why did the stock market drop after the announcement?
The announcement reinforced an existing fear: if AI agents absorb a portion of the work that justified per-seat software licences, the SaaS revenue model is under pressure. The IGV ETF has fallen more than 22% since the start of 2026, and approximately $2 trillion in market capitalisation has evaporated from the sector.
Which companies are seen as most exposed?
Those whose value proposition boils down to « we produce this deliverable faster », without a durable advantage: proprietary data, deep integrations, regulatory constraints, high switching costs. Salesforce (-28%), Atlassian (-35%) and Thomson Reuters (-16%) are among the hardest hit.
How did professionals react?
Through adoption and concrete testing. Harvey integrated Opus 4.6 and reported a 90.2% score on BigLaw Bench. Dentons is deploying Claude for legal drafting and research. Developers on Hacker News describe shifting from coding mode to review mode.
Does this mean the end of SaaS?
No. It means the end of a portion of easy rents. Companies that own proprietary data, business workflows and real operational accountability are more defensible than those selling an interface over work that has become automatable.
What is the relationship between Opus 4.6 and Sonnet 4.6?
Opus 4.6 (February 5) is the flagship model. Sonnet 4.6 (February 17) delivers performance comparable to Opus 4.5 at a fraction of the cost, becoming the default model for all free and Pro users.
What should marketing teams take away?
That model capabilities are advancing faster than most organisations’ ability to govern them. Differentiation is shifting to editorial framing, quality standards and AI workflow orchestration.