Anthropic released Claude Fable 5.2 and Claude Mythos 5.2 on Nov. 9, a week before the company is expected to begin the final stretch of its initial public offering. The new models are up to 30% faster than their predecessors and about 20% cheaper, with list prices of $8 per million input tokens and $40 per million output tokens. An optional fast mode costs $16 and $80 per million tokens and can generate output up to 2.5 times faster.
The timing is not subtle. Anthropic is asking public investors to value a company whose revenue is growing at an extraordinary rate and whose costs are growing almost as fast. The model launch is the product half of the pitch: faster, cheaper, and still at the frontier. The amended S-1 filing is the other half: a $567 million operating loss in the third quarter, a $9.9 billion compute bill, and $895 billion in long-term commitments. The IPO will test which half investors weight more heavily.
The models
Fable 5.2 and Mythos 5.2 share the same underlying model but ship with different safeguards. Fable 5.2 is broadly available on Anthropic's platforms, Claude Code, and the major cloud marketplaces, with stricter restrictions on cybersecurity and biology tasks. Mythos 5.2 has fewer restrictions and is limited to vetted organizations through Anthropic's Cyber and Life Sciences Verification Programs. The company still recommends Claude Opus 5.5 as its strongest daily driver for most coding and knowledge work, and says Fable and Mythos are "best for the most ambitious, open-ended technical work."
Anthropic says the new models can act as an advisor consulted by smaller models or as an orchestrator that delegates simpler work to other agents. In pre-release testing, the company says, they migrated software, worked on device design, and formalized mathematical proofs. The company also claims top spots on several benchmarks, including an Artificial Analysis Intelligence Index score of 65, though those results are company-reported and have not been independently replicated at scale.
The pricing is the more consequential number for enterprise customers. A 20% cut to the list price lowers the cost of running agents and long-context workloads, and the fast mode gives customers a way to pay for speed without switching models. The launch is a competitive answer to OpenAI's GPT-6 Astra and Google's Gemini 4 Carbon, and to cheaper Chinese models that have pressured prices across the industry.
The enterprise math is straightforward. A 20% cut on output tokens lowers the cost of agentic workflows, where models generate long outputs and call tools repeatedly. The fast mode gives customers a premium option for latency-sensitive work. Anthropic is betting that lower prices will expand usage faster than they reduce revenue per token. That is the same bet the rest of the industry is making, and it is why the IPO valuation depends on volume growth as much as on price.
The launch is also the first frontier release since Anthropic chief executive Dario Amodei called for slowing AI development in September 2026. The company's safety report says the new models are more likely to suspect they are being tested, though it still rates the risk of loss of control as low under its responsible scaling policy. An external evaluation by METR found that Anthropic's capability growth over the past 11 weeks ran roughly twice as fast as the pace from March 2024 to October 2025. The safety context matters for the IPO because investors are being asked to value a company that is arguing for caution while racing to stay at the frontier.
The IPO window
Anthropic filed its original S-1 on Oct. 26 and amended it in early November. The company plans to list on Nasdaq under the ticker ANT. The roadshow is expected around mid-November, with a listing targeted for the week of Nov. 16 or shortly before Thanksgiving, according to reports. The offering could raise up to $100 billion, which would make it one of the largest IPOs on record.
The valuation is the live question. Earlier expectations ran as high as $2 trillion. More recent reporting puts the target at roughly $1.5 trillion to $1.6 trillion, as investors weigh a choppy market for AI stocks and the company's mounting compute costs. A valuation at that level would still make Anthropic one of the most valuable companies in the world. It would also require public investors to accept a business that is growing quickly but losing money on an operating basis.
The market backdrop is not helpful. AI stocks have sold off as investors question the returns on massive data-center spending. The midterm elections gave Democrats control of Congress, which could mean more scrutiny of AI regulation, chip exports, and competition policy. Anthropic is trying to price its IPO into a market that is both excited about AI and worried about the cost of building it.
The model launch is designed to support that valuation. A company that is about to ask the public for capital wants to show that it is still at the frontier, that its products are getting cheaper to run, and that its customers have a reason to stay. The launch does all three. It does not answer the cost question.
The numbers in the filing
Anthropic's amended S-1 shows third-quarter revenue of about $17 billion, up roughly 12 times from a year earlier and 47% from the second quarter. The year-over-year growth is extraordinary. The quarter-over-quarter growth decelerated from 144% in the second quarter, a signal that the law of large numbers is starting to bite.
The operating result swung from a $974 million operating profit in the second quarter to a $567 million operating loss in the third. Adjusted operating income was $2.5 billion, a 15% margin, down from 19% in the prior quarter. The difference reflects the cost of compute. Infrastructure costs were $9.9 billion, or 58% of revenue, up from 48% in the second quarter, including $5 billion in short-term compute contracts that are more expensive than long-term capacity.
The balance sheet is strong but the commitments are enormous. Anthropic reported $895.4 billion in total purchase and lease obligations through 2045, including $447.5 billion due by the end of 2030 and $69.8 billion within 12 months. It had $146.6 billion in liquidity. The net loss was about $23.3 billion, largely because of non-cash fair-value charges on convertible notes and SAFE agreements; the nine-month net loss was about $104 billion, of which roughly $94 billion was non-cash.
The revenue quality also deserves scrutiny. About 44% of revenue was booked gross through cloud marketplaces, where Anthropic acts as a principal, and two customers accounted for about 20% of first-half revenue. Those are not fatal facts, but they are the kind of details public investors tend to notice.
The tension
The bull case is straightforward. Anthropic is growing faster than almost any enterprise software company in history, its models are at the frontier, and the compute commitments are a bet that demand will keep rising. The company says its models now generate 90% of its own code, which supports the argument that it can operate efficiently as it scales. If the AI market keeps growing, the compute bill is an investment, not a problem.
The bear case is equally clear. The company is losing money on an operating basis, its gross margin is under pressure from compute costs, and its growth is decelerating. The IPO valuation assumes that Anthropic can hold its pricing and its market share against OpenAI, Google, and cheaper Chinese models. The model launch is a reminder that the competition is intense and that price cuts are part of the strategy. A cheaper model is good for customers and bad for revenue per token.
The IPO will force investors to choose between those stories. The product launch says Anthropic is still winning at the frontier. The S-1 says winning at the frontier is expensive. Both can be true. The question is what multiple the market will pay for a company that is growing quickly but spending heavily to stay in the race.
The thread
This is the fifth part of the Anthropic IPO thread. The earlier pieces covered the company's decision to wait on its own bankers, its profit story, a $15 billion revolver, and $518 billion in compute commitments. The common thread was the gap between Anthropic's ambition and the capital required to fund it. The model launch adds a product chapter: the company is not only raising money to build compute, it is shipping products that it hopes will justify the spend.
The model launch closes one loop. The product is the pitch. The compute bill is the price of the pitch. The IPO is where the two meet.
What to watch
The roadshow will show how much demand exists at the target valuation. The pricing and the first-day trading will show whether public investors accept the AI growth story or apply a discount for the cost structure. The first post-IPO quarter will show whether the model launch translates into revenue and whether the compute costs stabilize.
Watch the customer concentration and the marketplace revenue mix. Watch the pricing pressure from OpenAI and Google. Watch the regulatory environment, which could change after the midterm elections. Watch the company's ability to convert its compute commitments into revenue without another round of price cuts.
The models are faster and cheaper. The IPO will decide whether that is a growth story or a cost story. Anthropic is asking investors to believe both. The first-day trading range will be the clearest signal of which story the market chooses, and the answer will set the tone for the next wave of AI listings.
Primary sources
- Reuters via The Star for the Fable 5.2 and Mythos 5.2 launch, pricing, and IPO context.
- Malay Mail for the model details and the $100 billion offering.
- SD Times and Constellation Research for the model capabilities, availability, safeguards, and the Anthropic quote.
- Business Insider for the amended S-1 financials and the revenue and loss figures.
- Investors.com for the Q3 results, compute costs, and valuation context.
- Asianet Newsable for the purchase commitments and the valuation range.