Corvex Launches a Security-Priority AI Inference Platform for Open-Weight Models
PR Newswire
44m ago
Ai Focus
Corvex announces the launch of Corvex Token Factory, a serverless inference platform aimed at open-weight AI models. By default, no data is retained, and access is provided through API. The company claims that this platform supports OpenAI and Anthropic compatible interfaces. The initial available models include Z.ai's GLM 5.3 and DeepSeek V4 Flash 0731.
Helpful
No.Help

Corvex Token Factory combines default zero-data retention with access based on API.

Arlington, Virginia, September 30th / PRNewswire / -- Engineering-driven AI Computing Company Corvex, Inc. (Nasdaq ticker: MOVE) today announced the launch of Corvex Token Factory. This is its serverless inference platform, designed to help developers and enterprises use open-weight AI models with greater data control capabilities.

Corvex indicates that Token Factory allows customers to access the open-weight AI model through API, without the need to deploy or operate a GPU cluster. The first batch of available models includes GLM 5.3 from Z.ai and DeepSeek V4 Flash 0731. This release also expands the service scope of Corvex to developers and enterprise technical teams who wish to use AI as a service.

Corvex, Co-CEO and Co-Founder, stated: "When companies use AI for source code, customer records, and internal documents, they need to know where this information will go and what will happen to it. With the help of Corvex Token Factory, we default to not storing end-users' prompts or responses, nor do we use any user data for training. We have built these control measures into a service that developers can easily use; they can leverage existing coding frameworks and AI API to securely enjoy the capabilities of leading open-weight models at a cost far lower than that of closed-source alternatives."

Inference is the process of converting end-user requests into responses generated by models. Corvex provides the hardware required to run each request and operates these models independently. Requests are not routed to third-party inference providers, nor are they forwarded to model developers.

At the beginning of its launch, Corvex Token Factory was built around four principles:

  • GuaranteeBy default, no data is retained. Prompt words and responses are processed in memory and are not recorded or stored, nor are they used for training models. Corvex retains a limited amount of operational metadata for security, service operation, and billing purposes, but this metadata does not include prompt words or response content.
  • Reliability– Corvex Token Factory is designed to maintain stability under continuous workloads, serving as a reliable execution layer in production workflows.
  • Performance– Corvex optimizes its inference stack to strike a balance between first-response time and high-load throughput, in order to support interactive applications and agent workflows.
  • ConcisenessAPI which is compatible with OpenAI and Anthropic allows developers to connect to existing clients by simply changing the basic URL, API keys, and model names. Corvex is responsible for the underlying infrastructure; the specific functional compatibility may vary depending on the client and model.

Corvex has passed the authentication of SOC 2 Type II. The company will also sign business partnership agreements with customers subject to the constraints of HIPAA, and the relevant security controls are recorded in Corvex Trust Center.

Customers pay based on the token they use, with no charges for idle GPU. Corvex Token Factory also serves as a starting point for more in-depth customer relationships. The roadmap for Corvex includes dedicated enterprise deployments and secure private inference environments, providing pathways for adopting additional services as customer usage and needs grow.

This release is conducted after the Corvex Token Factory closure of the alpha phase. Customers can register for free at the URL https :// tokenfactory.corvex.cloud / app, and will receive a API key. Product documentation is available at https :// docs.corvex.cloud.

Regarding Corvex, Inc.

Corvex, Inc. (Nasdaq ticker: MOVE) is a AI cloud computing company that specializes in AI workload-oriented GPU acceleration infrastructure. It is a publicly traded, pure AI computing platform. The company provides secure, scalable, and cost-effective computing resources through GPU accelerated clusters, high-throughput storage, and a layered architecture designed for scaled reliability, performance, and efficiency. Corvex's product portfolio includes AI Factories and GPU Clusters, Assured AI confidential computing platforms, as well as Corvex Token Factory serverless inference platforms. For more information, please visit corvex.ai.

Warning Regarding Forward-Looking Statements

This press release contains forward-looking statements as defined by the Private Securities Litigation Reform Act of 1995, which are based on current expectations or beliefs and assumptions about future events. Forward-looking statements include all statements that are not historical facts and can typically be identified by words such as “anticipate”, “believe”, “could”, “estimate”, “expect”, “intend”, “may”, “plan”, “potentially”, “seek”, “should”, “target”, “will”, or “would” and their negative forms. Such statements include, but are not limited to, those regarding product capabilities, customer deployments, business strategies, growth plans and future operational objectives, expected power supply and customer readiness dates, expected megawatt capacity and expansion options, as well as statements regarding expected customer purchase negotiations, contract terms, pricing, and additional sales opportunities. Due to various factors, actual results may differ significantly from these forward-looking statements. These factors include: the company's inability to reach any agreement regarding customer purchase agreements, or to do so within the expected timeframe, at the expected pricing, or in accordance with the take-or-pay terms; changes in the current market pricing for GPU service capacity; delays in the delivery of long-cycle equipment; delays in utility power connection or grid integration; local zoning, licensing, and suspension risks that affect any site expansion phase; the company's dependence on third-party site partners and the performance of formal agreements signed with them; the company's ability to obtain or maintain partner certifications and joint sales support; and the company's limited operating history.Due to the risks and uncertainties described under the "Risk Factors" section of SEC's report, as well as its most recent 10-K annual report and 10-Q quarterly report, actual results may differ significantly from those expressed or implied in the forward-looking statements. These documents can be found at www.sec.gov and in other documents submitted or to be submitted by the company to SEC. The forward-looking statements in this press release are only valid as of the date of issuance. Except as required by law, the company assumes no obligation to update or revise these forward-looking statements to reflect events or circumstances that occur after the date of this press release.

Media Contact

Chris Donahoe, Stillpoint

Tip
$0
Like
0
Save
0
Views 5
HQYC reminds readers to view blockchain rationally, stay aware of risks, and beware of virtual token issuance and speculation. All content on this site represents market information or related viewpoints only and does not constitute any form of investment advice. If you find sensitive content, please click“Report”,and we will handle it promptly。
Submit
Comment 0
Hot
Latest
No comments yet. Be the first!
Related
What happens to your tokenized assets if the issuer goes bankrupt?
A tokenized government bond can be transferred across chains in just a few seconds, and tokenized stocks can also be traded around the clock. However, when problems do arise, the most critical question is: if the company behind the tokens goes bankrupt, what exactly do you legally own? The answer depends less on blockchain and more on the legal structure behind those tokens.
Coinpaper
·2026-09-30 22:43:00
7
Tom Lee claims the crypto market has entered a bull market Bitmine approaches 5% ownership of all Ethereum
Tom Lee stated during the Korea Blockchain Week that the crypto market has entered a bull market phase, pointing to factors such as Bitcoin breaking through the 200-day moving average, institutional funds flowing back, the acceleration of tokenization, and the use of digital currencies by AI agents. Meanwhile, Bitmine claimed to hold 6,001,302 Ether, which accounts for approximately 4.9% of the total Ethereum supply.
Coinpaper
·2026-09-30 22:35:14
13
After a 40% rebound in stock price, Micron faces a financial test tonight: Net profit forecast surges by over 1000%, market closely watches next year's AI capital expenditure
Micron will release its financial report after the close of U.S. stocks on Wednesday. The market expects a significant increase in net profit and revenue for its fourth fiscal quarter, but investors are more concerned with management's assessment of the capital expenditures of hyperscale cloud providers next year, as well as whether the rise in memory prices will continue.
Wallstreetcn
·2026-09-30 22:35:12
14
The trading volumes of some products, such as Kalshi and Polymarket, have raised doubts, and they are under scrutiny amidst massive growth.
Some industry observers question that the trading volumes of certain products on prediction market platforms Kalshi and Polymarket may be exaggerated. The disputes focus on low-probability contracts, Ethereum perpetual contracts Kalshi, and whether there is wash trading. Both companies deny the existence of wash trading and claim that such patterns are not considered unnatural transactions.
CNBC
·2026-09-30 22:35:10
15
Bilibili opens-sourcees the Index-Translate multi-language translation model family; the text model supports 150 languages
Bilibili's Index LLM team officially released the Index-Translate model, with 2B, 9B, and 35B-A3B (preview) text model weights now available on Hugging Face and ModelScope. This model family is built upon Qwen3.5 and covers 150 languages, including Chinese and English, while also supporting translation instructions for terms, formats, and content preservation.
The Block
·2026-09-30 22:23:11
18
View More