Vice President Finance

DeepSeek ships a faster Flash model and its own search API. DeepSeek is sticking to its playbook: little fanfare, a lot of performance per cent. On Saturday, the Chinese A.I. lab released DeepSeek-V4.1-Flash-1017, a new version of its fast Flash model. According to DeepSeek's own changelog, it delivers better benchmark results, finishes tasks faster and produces shorter answers. Prices stay the same. DeepSeek is also opening its in-house search engine to developers for the first time. More performance for the same price. The update is already live in the web app, the mobile apps and the API, under the name deepseek-flash. The model makes its biggest gains in coding and in so-called agentic tasks, in which the A.I. completes several steps on its own. On Terminal-Bench 4.0, its score rises from 31.2 for the September version to 42.3, and on the DeepSWE software engineering test from 74.2 to 77.0. The model was optimized mainly for DeepSeek Harness, the company's own agent environment, and for Tencent's coding assistant CodeBuddy. Under the hood, V4.1 Flash is a mixture-of-experts model with 552 billion parameters, of which only 8 billion to 16 billion are active per token. Its context window holds one million tokens. Pricing remains $0.30 (about €0.27) per million input tokens and $1.20 (about €1.07) per million output tokens. Because answers are shorter, tasks should end up cheaper overall. A search engine as an API. The search API is new. DeepSeek runs its own search engine, which the company says indexes hundreds of billions of web pages and handles hundreds of millions of queries a day. It is designed to filter out SEO spam, GEO manipulation and low-quality A.I. content. Through the API, developers can now retrieve search results or have individual web pages returned as text. A thousand successful requests cost $0.75 (about €0.67). That puts DeepSeek in competition with providers such as Brave, Exa and Tavily, whose search APIs power many A.I. agents. Where deepseek ranks among open-weight models. The independent benchmarking site Artificial Analysis has not yet tested the new version. The September version of V4.1 Flash scores 39 points on its Intelligence Index. Among models with openly available weights, that puts DeepSeek in a tie for sixth place with Meta's Muse Spark Open. All of the models ahead of it come from Chinese rivals: * No. 1: Xiaomi's MiMo-V2.6-Pro, with 46 points * No. 2: Z.ai's GLM-5.3, with 45 points * Nos. 3 and 4: StepFun's Step 5 Preview and Moonshot's Kimi K3, with 44 points each * No. 5: Z.ai's GLM-5.3-Flash, with 42 points In the overall ranking of all models, DeepSeek lands only around 46th place. The top spots belong to closed models from Anthropic, Meta, OpenAI and Google, with up to 58 points. Meta is the strongest American provider of open-weight models, followed by Reflection with Beam 1.0 (30 points) and Nvidia with Nemotron 3.5 Super VL (26 points). An analyst at the routing provider OrcaRouter predicts 45 to 46 points for the new version, which would be enough for the top three. That has yet to be measured, however. Robert Lea, an analyst at Bloomberg Intelligence, estimates that the gap between Chinese models and the leading American ones has narrowed to about 3 percent since the V4.1 launch, down from 15 percent earlier this year. The numbers come from DeepSeek. The benchmarks call for caution. All comparison figures for the new version come from DeepSeek itself, and in several tests the model still trails the expensive top models from the United States. It is also unclear whether DeepSeek will publish the weights of the new checkpoint. The September version is freely available on Hugging Face under an MIT license, but reports on the update are contradictory. Companies that build products on DeepSeek also have to live with silent updates: The model behind the name deepseek-flash changes without much notice. Founders Dispatch Europe's Startup Scene, Twice a Week in Your Inbox. Deals, funding rounds and tech trends, curated by the Trending Topics newsroom. Free since 2015. Aus Datenschutz-Gründen ist dieser Inhalt ausgeblendet. Die Einbettung von externen Inhalten kann in den Datenschutz-Einstellungen aktiviert werden:

Pyth and Exa explore new Market Data model for AI agents. 9 October 2026 Pyth Network and AI-native search provider Exa are collaborating to explore how AI agents can combine financial market data with web-based research, potentially opening up new approaches to data distribution, licensing and consumption as agentic AI becomes more widely adopted in financial services. The collaboration, currently in an early testing phase, brings together Exa's search and agent infrastructure with Pyth's structured market data, including real-time prices, historical observations and candlestick data. Through Pyth's APIs and Model Context Protocol (MCP) server, agents can retrieve market information alongside research gathered from the web, combining the two within a single workflow. Initial applications include event-driven financial research, historical market analysis, cross-asset comparisons and portfolio monitoring. An agent investigating an earnings announcement, for example, could retrieve relevant news and company information, identify the appropriate financial instrument and analyse its price movements over subsequent trading sessions. In conversation with Market & Alt Data Insight, Michael Cahill, CEO of Douro Labs and co-founder of Pyth Network, sees the collaboration as an opportunity to explore some of the challenges facing established market-data distribution models as financial applications become increasingly autonomous. Rethinking Market Data Distribution Traditional market-data infrastructure has evolved around relatively predictable consumption patterns, with financial institutions typically acquiring data through established vendor relationships, exchange agreements and application-specific licences. While existing platforms can accommodate AI capabilities within their environments, Cahill believes companies building AI-native applications from the ground up have different requirements, particularly when it comes to sourcing and integrating information across multiple asset classes. "The big problem with the market-data sector is that it's a myriad of different vendors that you have to interact with to get a complete picture," observes Cahill. "I liken it to cable television back in the 1990s. You would have a cable operator, and then you'd have channels that came as part of your bundle, and premium channels you'd have to contract for separately. That's more or less how financial market data works today. You've got FactSet, Bloomberg and Refinitiv as the brokers, and then you buy data directly from exchanges. You have to go around the world to get a complete picture." Pyth's approach centres on a network of financial institutions contributing pricing information, including trading firms such as Jump Trading, Jane Street, Hudson River Trading and DRW. The company The resulting Pyth-native data is made available through a common interface, reducing the need for applications to integrate multiple data feeds independently. For an AI-native company such as Exa, this provides a way to incorporate structured financial data into its search infrastructure without having to establish separate integrations and commercial arrangements across numerous providers. The Licensing Challenge Existing market-data licensing arrangements generally distinguish between internal use, display, redistribution and other categories of consumption. AI introduces additional questions about how data is accessed and used, particularly when agents dynamically retrieve information, combine multiple sources and generate outputs without following predetermined application workflows. "If you want to consume data from an internal-use perspective, that's very non-controversial. As soon as you start getting into rights, that's where the market ends up being the slowest and the stickiest," notes Cahill. "If you run a large data business, you probably don't know how to price inference, training, display and various other things that are now novel with AI consumption and usage. You're going to be apprehensive, and that's what we've seen the industry doing. This is where we can be more aggressive." The Exa pilot will help Pyth understand how agents consume financial data, including the frequency, volume and nature of their requests, and use those findings to develop an appropriate commercial model. Unlike traditional data distributors that must negotiate rights with multiple underlying suppliers, Pyth argues that its contributor-based model gives it greater flexibility over how its native data can be licensed. That flexibility could become increasingly valuable if agentic workflows generate substantially different usage patterns from conventional terminals and applications. The pilot has yet to establish which pricing structures will prove commercially viable, however. Provenance and Trust in Agentic Workflows Agents combining structured market observations with unstructured research need sufficient context to interpret information correctly, particularly when the underlying sources operate on different timescales or use different identifiers. An agent analysing an earnings announcement, for example, needs to distinguish between the announcement's publication time, the relevant trading session and the period over which a subsequent price movement should be measured. It must also identify the correct security, venue and currency where multiple possibilities exist. The companies argue that this information needs to be explicitly machine-readable rather than relying on the interpretation that a human analyst might bring to a conventional market-data terminal. Pyth's infrastructure was originally developed to provide cryptographically secured pricing information for blockchain-based financial transactions, where incorrect data can have immediate and potentially irreversible consequences. "That's where I think our legacy of doing this for blockchain and crypto markets - dealing with machines, but with value at risk - gives us a huge advantage over something that has no comparable stakes, or is primarily designed for entertainment or display," says Cahill. "What we designed was a system that could withstand the fragilities of a standard publishing paradigm, while also making sure it was useful from a real-time perspective. We provide an easy-to-use source where agents don't have to make all these inferences and decide whether the data can be trusted." Within the Exa collaboration, Pyth provides structured market observations and associated data context, allowing Exa's search infrastructure to concentrate on retrieving, interpreting and synthesising information from other sources. For institutional applications, maintaining a distinction between what an external source reports, what the market data records and what the AI model subsequently infers will be important, particularly where research processes need to be transparent and auditable. Beyond Financial Research Event-driven research and historical market analysis are the most immediate applications being explored, but Cahill envisages agents progressing from retrieving information and analysing market events towards constructing investment strategies and eventually initiating transactions autonomously. He connects that development with the growth of what he calls everything exchanges - trading platforms offering access to an increasingly broad range of financial exposures, including through perpetual derivatives. Broader cross-asset pricing coverage, increasingly continuous trading and autonomous software could eventually make it easier for agents to manage capital across different markets. Such applications would also introduce additional requirements around execution controls, governance and regulatory oversight. For Pyth, AI-native companies that have yet to establish extensive relationships with incumbent financial-data providers represent a potential new customer base. Cahill believes these businesses may be more receptive to alternative sourcing and licensing arrangements than established institutions with substantial investments in existing infrastructure. The Exa collaboration will provide an early indication of whether Pyth's approach can meet those requirements. The companies are still testing consumption patterns and exploring commercial arrangements, with no production-wide deployment announced. Whether the model gains traction will depend on its ability to deliver reliable financial data at a cost and under licensing terms that make sense for AI-native applications.

Cloudflare announces Web Search API integration for AI Gateway in partnership with Ceramic.ai, Exa, and Linkup. 6 hrs ago - 6 October 2026 Siti Muinah OvFile Reporter Cloudflare has announced a major expansion of its artificial intelligence infrastructure, introducing a new Web Search API designed to bridge the gap between static AI models and real-time internet data. By integrating live web search directly into its AI Gateway, the company aims to solve a fundamental inefficiency in how autonomous AI agents interact with the web, while simultaneously setting a new standard for transparent and ethical bot crawling. The initiative launches with foundational support from prominent web search providers Ceramic.ai, Exa, and Linkup. This collaboration is designed to equip developers with the tools necessary to ground AI responses in live information, moving away from outdated knowledge cutoffs and erratic URL-guessing behaviors. Overcoming the inefficiencies of AI web access. To understand the significance of the new Web Search API, industry experts point to a common, albeit inefficient, practice currently employed by many AI agents. When an agent is tasked with retrieving information from a live web page, it typically attempts to guess the target URL before making a tool call to execute a curl command. This speculative approach frequently results in errors, most notably the dreaded 404 Not Found response, triggered when the agent's algorithmic guess misses the mark. Relying on random URL enumeration is not only computationally wasteful but also hinders the reliability of AI applications that require precise, up-to-date data. Recognizing this limitation, Cloudflare's new offering introduces a more deliberate mechanism. By mirroring the human approach to information retrieval - starting with a search engine query - the Web Search API enables agents to query the internet for relevant data systematically. This injects fresh, structured snippets directly into the model's context layer, grounding its responses in real-time information rather than relying solely on pre-trained parameters. Addressing the knowledge cutoff challenge. Artificial intelligence models are traditionally constrained by their training data, which is frozen at a specific point in time. This knowledge cutoff date creates significant barriers for users attempting to engage models in discussions regarding recent events, shifting application programming interfaces, or fast-evolving news cycles. By embedding the Web Search API directly into the inference pipeline, developers can establish a dynamic context layer for their applications. Rather than operating in a historical vacuum, models can access live information seamlessly. For instance, developers building applications utilizing Cloudflare developer tools often encounter challenges keeping pace with rapid product updates and feature rollouts, such as those announced during Cloudflare's annual Birthday Week celebrations. With the new integration, developers and their AI agents can instantly retrieve the latest documentation and release notes, facilitating faster and more accurate application development. Elevating industry standards for responsible web crawling. Beyond technical efficiency, Cloudflare's announcement places a strong emphasis on the ethical dimensions of web crawling and data consumption. The company has long advocated for transparency, honesty, and strict adherence to established bot rules and preferences, asserting that website owners must maintain meaningful control over how their digital content is utilized. In alignment with these principles, all launch partners for the Web Search API have committed to meeting Cloudflare's rigorous bot crawling standards. Under this agreement, the crawlers utilized by search providers must strictly comply with Cloudflare's publicly stated requirements for verified bots, as outlined in the platform's developer documentation. Furthermore, any web search responses generated through the API must include a direct link to the original location of the crawled content. Industry observers note that these rules represent a positive step toward maintaining a fair internet ecosystem. By requiring crawlers to identify themselves, respect robots.txt protocols, and properly attribute sources, the partnership ensures that content creators retain visibility and control over their intellectual property while still participating in the growing AI economy. Deep integration with AI Gateway. The new Web Search API is anchored within Cloudflare's AI Gateway, a flagship control plane designed to provide unified billing, observability, security, and access controls for AI applications. By routing web search through this existing infrastructure, developers can manage their AI workflows within a single centralized dashboard. Requests made to the Web Search API are automatically logged alongside standard AI Gateway observability data, and search queries draw directly from the user's existing AI Gateway credit balance. To address data privacy concerns, Cloudflare has committed to identifying partners that support Zero Data Retention policies, ensuring that sensitive information is not stored by search providers. Additionally, the service is offered directly at the partners' list API prices without any added platform markups. The infrastructure also accommodates flexible integration patterns. Organizations utilizing Bring-Your-Own-Key configurations for model inference can similarly apply their existing organizational setups to web search providers. For backend systems, mobile applications, and external services, developers can utilize a standard REST endpoint by passing an AI Gateway authentication token and specifying their preferred provider in the payload. For developers building natively on Cloudflare Workers, the integration requires minimal code through a dedicated worker binding. This allows agents to execute search queries and process results with high performance and low latency. Looking ahead, Cloudflare is developing native Server Tools directly within AI Gateway, which will soon eliminate the need for developers to manually define tools and orchestration harnesses, with web search slated as one of the first built-in offerings. The Web Search API is available now for developers seeking to enhance their AI agents with grounded, real-time intelligence. Interested parties can access implementation details through the official Cloudflare developer documentation or test the capabilities directly within the AI Playground using an active AI Gateway account and key. Related topics.