Factory Sells Model Switching, and Now Anthropic Commitments Can Pay for It

From $1.5 billion to $5 billion in five months

Factory, which builds AI coding agents for enterprises, announced on Sept. 15 that it had raised $200 million at a $5 billion valuation. Blackstone, Khosla Ventures, Sequoia Capital, Insight Partners, Evantic Capital, Sound Ventures, NEA, Mantis VC and Clearlake took part, bringing total funding to more than $400 million. The company said it will put the money into research, product and global go-to-market.

According to TechCrunch, Factory raised $150 million at a $1.5 billion valuation in April in a round led by Khosla Ventures. The valuation more than tripled in five months. Matan Grinberg and Eno Reyes founded Factory in 2023. Grinberg, then a PhD student at UC Berkeley, cold-emailed Sequoia’s Shaun Maguire, which led to the company, and Sequoia backed it from the seed stage.

The customer list has changed too. The April TechCrunch piece named engineering teams at Morgan Stanley, Ernst & Young and Palo Alto Networks. This announcement names Nvidia, Blackstone, Royal Bank of Canada, Palo Alto Networks, Adobe and T-Mobile. Factory says hundreds of thousands of developers use its product, and it emphasizes that it can run in the cloud, on premises, or in fully air-gapped environments.

A checkout line funded by Anthropic commitments

On Sept. 9, six days before the funding news, Factory joined Claude Marketplace. Anthropic opened the marketplace in March so enterprise customers can apply part of their committed Anthropic spend to Claude-powered partner products. Partner purchases count against a portion of the existing commitment, and Anthropic handles all invoicing. The launch partners were GitLab, Harvey, Lovable, Replit, Rogo and Snowflake.

Factory’s announcement framed the benefit as skipping a separate buying cycle. A company that already has a commitment with Anthropic does not need to start procurement from scratch to buy Factory. At launch, The Next Web reported that Anthropic was comparing the model to the AWS and Azure marketplaces, which let customers redirect existing cloud commitments to partner tools. Unlike those, Anthropic was not taking a commission, at least at launch.

A model-agnostic company inside one model maker’s ledger

The listing stands out because of what Factory has pitched as its edge. According to TechCrunch, Grinberg told The Wall Street Journal in April that Factory’s key differentiator is its ability to switch between foundation models such as Anthropic’s Claude or China’s DeepSeek. Factory has also launched Factory Router, which assigns models automatically by task: frontier models handle planning and hard diagnosis, while lighter models take on search and targeted checks.

In coding tools, Anthropic is both supplier and competitor. The April TechCrunch piece listed Anthropic, maker of Claude Code, alongside Cursor and Cognition as companies vying for the market. The Next Web also noted at the marketplace launch that Anthropic’s own enterprise products overlap at the margin with partner tools. Replit and GitLab were among the launch partners.

According to Factory, companies using connected deployments can run Claude through their approved Anthropic setup while keeping execution close to their code and infrastructure. Anthropic’s marketplace lead described Factory as now purchasable with the commitment customers already have with Anthropic.

The savings number changes from document to document

How much routing saves depends on which Factory document you read. The funding announcement says token spend fell by more than 60%. The partner network announcement says routing can cut token costs by more than 50% at the same quality. A technical post on routing says that after more than two months on customer work, aggregate cost fell 58% compared with using the frontier model for every call. The same post puts the median session saving at 76% and says median turn latency dropped from 81 seconds to 49.

Only the 58% figure states its baseline, and that baseline is a frontier model for every call. A team that already splits work across models may see smaller savings from the same tool. The same post notes that prices among models capable of serious agent work span more than fiftyfold, so the choice of baseline can swing the savings figure a long way.

Cutting cost is one question; what the spend produces is another. When Factory launched Agent Effectiveness in August, it wrote that engineering leaders are spending more on AI but still cannot answer what that spend is producing. The feature links Factory sessions to records in issue tracking and source control tools to show which work and artifacts AI spend led to.

Tokyo comes first in Asia

On Aug. 31, Factory said it would open an office in Tokyo and expand across Asia. It hired Seiji Sasaki as President and GM of Japan. Sasaki previously led go-to-market and served as site lead for OpenAI Japan, and before that was VP and GM for Japan and Korea at Slack. Factory is hiring for sales, marketing, support and partnerships in the region. The announcement does not mention a Korea office.

For founders selling enterprise AI products in Asia, two shifts overlap. A US coding agent company is going after customers in the region from a Tokyo base, and model makers’ committed budgets are turning into a sales channel. Claude Marketplace says it is looking for Claude-powered products built for enterprise security, scale and compliance needs, and partners apply through a waitlist. Whether a prospect already has an Anthropic commitment is now a fair question for a first sales meeting.