Anthropic : $965 billion confidential IPO and Opus 4.8 propelled to the top
🔎 The day AI crossed the trillion-dollar mark
On June 1, 2026, Anthropic filed a confidential filing with the SEC for its public listing. A technical move, but one with massive implications: the company founded by Dario Amodei is targeting a valuation of $965 billion, just shy of a trillion.
This is the first time an AI company has reached this level prior to its IPO. Above all, Anthropic beats OpenAI in the race to go public, a scenario no one anticipated eighteen months ago.
The timing alongside the release of Claude Opus 4.8 on May 28 is no coincidence. A record model, a record raise, a record IPO filing. Anthropic is stringing together moves to maximize pressure on its competitors and win over institutional investors.
One question remains: is a valuation close to a trillion dollars justified, or are we witnessing the biggest financial gamble of the decade?
The essentials
- Anthropic files a confidential S-1 prospectus with the SEC on June 1, 2026, preparing a potentially historic IPO according to Wedbush (CNBC).
- The Series H, closed in late May, raises $65 billion at a post-money valuation of $965 billion, co-led by Altimeter, Sequoia and Coatue (Sherwood News).
- Annualized revenue reaches $47 billion, a pace that partly justifies the valuation (Enoumen).
- Claude Opus 4.8 reaches 69.2% on SWE-Bench Pro, 83.4% on OSWorld-Verified and 1890 Elo on GDPval-AA (Anthropic, ForkLog).
- Anthropic surpasses OpenAI in valuation for the first time, reversing the AI market hierarchy (AP News).
Recommended tools
| Tool | Main usage | Price (June 2026, check on anthropic.com) | Ideal for |
|---|---|---|---|
| Claude Opus 4.8 | Development agent, complex reasoning | 5$ / 25$ (input/output per million tokens) (Lushbinary) | Engineering teams with agentic workflows |
| Claude Code | IDE with Dynamic Workflows and parallel subagents | Included in Enterprise plans | Large-scale multi-file development |
| Claude Opus 4.7 (Adaptive) | Adaptive general reasoning | Quote-based | Mixed use cases requiring flexibility |
Series H: $65 billion in a single raise
Anthropic just closed the largest private funding round in tech history. $65 billion in a Series H, at a post-money valuation of $965 billion.
The round is co-led by Altimeter, Sequoia, and Coatue, with the participation of several institutional funds. According to InfoBro AI, this is highly likely the final private raise before the IPO.
The signal is clear: the most demanding investors in the market are betting that Anthropic can support a valuation close to a trillion dollars. This is not a speculative bet on a prototype. It's a bet on a business with $47 billion in annualized revenue, according to Enoumen.
To put this into context: at $965 billion, Anthropic is worth more than the market capitalization of many S&P 100 companies. And this is before even going public.
The timing is calculated. Anthropic lève 65 milliards de dollars en Series H et lance Claude Opus 4.8 : le jour où tout a changé marks a tipping point in market perception.
Why such a massive raise before the IPO?
Two reasons. First, to bolster the war chest to fund the compute infrastructure necessary for training next-generation models. Second, to lock in the private valuation at $965 billion so that the IPO pricing starts from a high floor.
It's a classic tactic, but at this scale, it is unprecedented.
The Confidential IPO Filing: Strategy and Signals
On June 1, 2026, Anthropic confidentially files its S-1 prospectus with the SEC. This mechanism allows the company to discuss with regulators without making its finances public immediately.
According to Yahoo Finance, Anthropic is deliberately outpacing OpenAI in the IPO race. The first AI lab to go public will capture the bulk of media attention and institutional flows.
Dan Ives of Wedbush, quoted by ABC7, speaks of "opening the floodgates for the dormant IPO market." The IPO market had been sluggish since 2022. An IPO nearing a trillion could reignite the entire cycle.
What the confidential filing tells us
A confidential filing means Anthropic is not sure of the exact IPO date. It could take 3 months or 9 months. But mostly, it means that auditors and investment banks have validated that the numbers hold up.
The WSJ, relayed by Techmeme, confirms that Anthropic has now surpassed OpenAI in valuation. A historic reversal for a company that, in 2024, was still perceived as a serious but distant challenger.
The stakes go beyond Anthropic. If this IPO succeeds, it validates the thesis that generative AI deserves valuations in the order of a trillion. If it fails or disappoints, the entire sector will take a hit.
Claude Opus 4.8: the model that justifies the valuation
On May 28, 2026, two days before the Series H announcement, Anthropic releases Claude Opus 4.8. The timing is not coincidental: they need to show investors that the product pipeline remains dominant.
Opus 4.8 introduces two major innovations: Dynamic Workflows in Claude Code and effort control. Dynamic Workflows allow the agent to delegate subtasks to parallel subagents, autonomously solving very large-scale problems.
The benchmark figures are impressive, but deserve nuanced analysis.
Benchmarks: where Opus 4.8 dominates
| Benchmark | Claude Opus 4.8 | Best competitor | Source |
|---|---|---|---|
| SWE-Bench Pro | 69.2% | — | ForkLog |
| Humanity's Last Exam (no tools) | 49.8% | — | ForkLog |
| Humanity's Last Exam (with tools) | 57.9% | — | ForkLog |
| OSWorld-Verified | 83.4% | — | ForkLog |
| GDPval-AA (Elo) | 1890 | — | ForkLog |
On real-world software development (SWE-Bench Pro, OSWorld-Verified), Opus 4.8 is clearly above the rest. It is the most capable model for solving complex GitHub tickets end-to-end.
Where Opus 4.8 does not dominate
On Terminal-Bench 2.1, Opus 4.8 reaches 74.6%, but OpenAI's GPT-5.5 leads with 78.2% according to Digital Applied. This benchmark specifically measures the ability to reliably execute tasks in a terminal.
This is an important point: no model dominates everywhere. The landscape remains competitive, and Anthropic's advantage is concentrated on agentic software development, not on all reasoning tasks.
Fast mode and pricing: faster, cheaper
Opus 4.8 includes a fast mode that is 2.5x faster and costs 3x less than previous Opus models, according to Anthropic. The official pricing is $5 per million input tokens and $25 for output.
It's not cheap, but it is in line with GPT-5.5 in the premium segment. Anthropic's selling point: for agentic development, Opus 4.8 solves problems in fewer steps, so the total cost per task is often lower despite a higher per-token price.
Anthropic vs OpenAI: The Reversal of the Balance of Power
For years, OpenAI dictated the pace. GPT-4 in 2023, GPT-5 in 2025, then GPT-5.5. Anthropic followed with a lag of a few months.
In June 2026, the dynamic changed. Anthropic is ahead on three simultaneous fronts: valuation, the race to IPO, and agentic development benchmarks.
The comparison of agentic scores from the Artificial Analysis Intelligence Index is revealing:
| Model | Agentic Score | Publisher |
|---|---|---|
| GPT-5.5 | 98.2 | OpenAI |
| Gemini 3 Pro Deep Think | 95.4 | |
| Claude Opus 4.7 (Adaptive) | 94.3 | Anthropic |
| GPT-5.4 Pro | 91.8 | OpenAI |
| o1-preview | 90.2 | OpenAI |
Opus 4.7 is close on the heels of GPT-5.5 on the overall index. But in specific development (SWE-Bench Pro), Opus 4.8 takes the lead. The nuance is essential: being the best general model is no longer enough. Companies buy for specific use cases, and Anthropic is winning on dev.
The article Anthropic vise 900 milliards de dollars : le round de 30 milliards qui dépasse OpenAI traces the trajectory that led to this reversal.
What the valuation actually means
At 965 billion with 47 billion in annualized revenue, the multiple is around 20x. For a growth company in tech, this is high but not absurd. In its early days, AWS sported similar multiples.
The risk isn't the multiple. The risk is the sustainability of the revenue. If companies cut their AI budgets or if competition (DeepSeek, open-source models) erodes prices, the multiple will collapse.
Dynamic Workflows : the true innovation of Opus 4.8
Benchmarks make the headlines, but the most strategic feature of Opus 4.8 lies elsewhere. Dynamic Workflows in Claude Code change the very nature of what a model can accomplish.
Until now, an AI agent worked sequentially: read the code, modify a file, test, repeat. With Dynamic Workflows, Claude can create sub-agents that work in parallel on different parts of a problem.
What does this change in practice?
An example: a ticket asks to refactor an authentication module, update the tests, and modify the documentation. Instead of doing these three tasks one after the other, Opus 4.8 can delegate each task to a subagent, execute them in parallel, and then merge the results.
The time saving is not linear. It is multiplicative on large-scale projects. This is what explains the score of 69.2% on SWE-Bench Pro: the benchmark measures the resolution of real tickets, where the ability to work in parallel is a decisive advantage.
Effort control also allows you to adjust the "depth of thought" of the model. For a simple bug, fast mode. For an architecture problem, deep think mode. The user pays for what they actually consume.
Corporate joint ventures: the engine behind $47B in revenue
A $965 billion valuation requires substantial revenue. Anthropic announced an annual revenue run rate of $47 billion, a figure that surprised many analysts.
A significant portion of this revenue comes from large-scale enterprise deployments. The business model has evolved: Anthropic no longer just sells tokens via API. It sells complete solutions to enterprises, including the model, infrastructure, and support.
The article Anthropic et OpenAI lancent chacun leur JV entreprise : 10 milliards de dollars pour déployer l'IA dans les PME et grands groupes details this dynamic. Both giants have understood that the real battle is won in enterprises, not with individual developers.
The business model behind the figure
Anthropic's enterprise contracts are structured around three components:
- Annual license for access to Claude via a private and secure interface.
- Token consumption billed beyond an included threshold.
- Professional services for integration into existing workflows.
This hybrid model (SaaS + usage) generates more predictable and higher revenue per client than simple API billing. This is also what makes the 20x multiple more defensible: it is not volatile transactional revenue, it is multi-year contracts.
An IPO at ~1 trillion: what it means for the market
If Anthropic goes public at a valuation close to a trillion dollars, the consequences extend far beyond the company itself.
Firstly, it creates a pricing precedent for all AI companies. If the market accepts 965 billion for Anthropic, then the private valuations of xAI, Mistral, and even OpenAI's future raises will be calibrated against this anchor point.
Secondly, it attracts new institutional investors to the sector. Funds that only invest in publicly traded companies will finally be able to expose their portfolios to generative AI. The spillover effect on the stocks of Nvidia, Microsoft, and Alphabet would likely be positive.
Thirdly, it increases regulatory pressure. A publicly traded AI company valued at a trillion dollars with a model that can write code autonomously will draw increased attention from US and European lawmakers.
The specific risks of this IPO
The main risk is dependence on the innovation cycle. If Opus 4.9 or 5.0 fails to maintain the competitive edge, revenue could stagnate. And a 20x multiple doesn't forgive stagnation.
The second risk is client concentration. Anthropic has not publicly detailed the breakdown of its 47 billion in revenue. If 30-40% comes from 5-6 major clients, the loss of a single contract could be devastating for the post-IPO valuation.
The third risk, less discussed, is the cost of infra. Each new generation of model costs hundreds of millions, if not billions, in compute. Anthropic's operational profitability remains a question mark that the S-1 will make public.
The competitive landscape: Google, xAI, and DeepSeek
Anthropic is not playing alone. The landscape of June 2026 is more competitive than ever.
Google positions Gemini 3 Pro Deep Think as the most powerful reasoning model, with an agentic score of 95.4. And Gemini 3.1 Pro dominates the overall ranking with 92 points. Google's advantage: native integration in Search, Cloud, and Android. An ecosystem that Anthropic does not have.
OpenAI remains the absolute leader on the agentic index with GPT-5.5 at 98.2. And GPT-5.5 leads on Terminal-Bench 2.1 (78.2% vs 74.6% for Opus 4.8). OpenAI also has the advantage of consumer brand awareness via ChatGPT, with hundreds of millions of users.
xAI with Grok 4.1 reaches 90 overall and 79 in agentic. Not the leader, but competitive, with the advantage of the X (Twitter) ecosystem for real-time training data.
DeepSeek is the wildcard. DeepSeek V4 Pro (Max) reaches 88 overall, a remarkable score for a Chinese model. If geopolitical tensions intensify, Western companies might be forced to choose between Anthropic/OpenAI and alternatives, which indirectly reinforces the American duopoly.
Premium model pricing comparison (June 2026)
| Model | Input price (per M tokens) | Output price (per M tokens) | Agentic Score |
|---|---|---|---|
| Claude Opus 4.8 | 5$ | 25$ | Not ranked (new) |
| GPT-5.5 | ~6$ | ~30$ | 98.2 |
| Gemini 3 Pro Deep Think | ~4$ | ~20$ | 95.4 |
| Claude Opus 4.7 | 5$ | 25$ | 94.3 |
| GPT-5.4 Pro | 4$ | 18$ | 91.8 |
Indicative prices, please verify on the respective websites.
Anthropic positions itself in the high-end segment, but not above the market. The pricing strategy is not a differentiator: it is the performance on dev tasks that justifies the premium.
❌ Common mistakes
Mistake 1: Confusing valuation with cash in the bank
The $965 billion valuation does not mean Anthropic has $965 billion in cash. The Series H raised $65 billion. The rest is a theoretical valuation based on expected future revenues. Commentators who talk about Anthropic as a "trillion-dollar company" exaggerate the financial reality.
Mistake 2: Comparing the Anthropic IPO to the Meta or Google IPOs
Meta (2012) and Google (2004) were profitable with high margins before their IPOs. Anthropic has $47 billion in revenue but its net profitability is not yet known. The S-1 will reveal whether the company is profitable or if it is still burning cash for training.
Mistake 3: Believing that Opus 4.8 dominates on all benchmarks
It does not. GPT-5.5 leads on Terminal-Bench 2.1 (78.2% vs 74.6%). Gemini 3.1 Pro generally leads (92 vs unpublished score for Opus 4.8 generally). Choosing a model should be done by use case, not by an overall score.
Mistake 4: Ignoring post-IPO regulatory risk
A publicly traded ~$1 trillion AI company will be under the microscope of US senators, the FTC, and the European Commission. Transparency obligations increase drastically after an IPO, which could slow down the pace of releasing new models.
❓ Frequently Asked Questions
Is Anthropic really worth more than OpenAI?
In private valuation, yes: $965 billion compared to an estimated $800-850 billion for OpenAI in May 2026. But OpenAI potentially generates more consumer revenue via ChatGPT. The comparison depends on which metrics you prioritize: valuation, revenue, or enterprise market share.
When will Anthropic's IPO take place?
The filing is confidential, so no date is public. Wedbush analysts estimate a possible launch between late 2026 and early 2027, depending on market conditions and the progress of discussions with the SEC.
Is Claude Opus 4.8 better than GPT-5.5?
It depends on the task. On SWE-Bench Pro (software development), Opus 4.8 leads. On Terminal-Bench 2.1 (terminal tasks), GPT-5.5 leads. On the overall agentic index, GPT-5.5 remains in the lead (98.2 vs a lower estimate for Opus 4.8). Neither model dominates everywhere.
Which investors participated in the Series H?
Altimeter, Sequoia, and Coatue co-led the $65 billion round. Other institutional funds participated, but the full list has not been made public.
Are Dynamic Workflows available to everyone?
Dynamic Workflows are integrated into Claude Code, Anthropic's agentic IDE. They are accessible via enterprise plans and volume contracts. Exact availability for individual developers depends on the subscribed plan.
✅ Conclusion
In May-June 2026, Anthropic followed up a $965 billion Series H with the release of Claude Opus 4.8 and its revolutionary Dynamic Workflows, along with the confidential filing for its IPO. The company has reversed the balance of power with OpenAI and is laying the foundations for an IPO that could redefine the tech market. It remains to be seen whether $47 billion in revenue and cutting-edge benchmarks will be enough to justify a trillion-dollar valuation in the eyes of public investors.