July 2026 Summaries

Merge introduces Fusion, a novel approach to AI model integration that leverages a panel of models to deliver superior answers at a fraction of the cost of frontier models. Fusion operates by sending a prompt to multiple models, each working independently, and then synthesizes their outputs using a judge model to create a unified response. This system capitalizes on the varying insights from different models, even when they agree, to enhance the quality of the final answer. Fusion has demonstrated exceptional performance on the DRACO benchmark, outperforming individual models and offering frontier quality at significantly reduced costs. It is particularly advantageous for tasks where answer quality is prioritized over latency, such as deep research and high-stakes decision-making in fields like legal and medical compliance. While it may not be suited for real-time, high-volume applications, Fusion provides a cost-effective alternative to single frontier models, ensuring continuity of service even if individual providers face disruptions.

AI governance platforms are crucial for companies as they integrate AI tools with internal systems, providing control over access, preventing data leaks, and maintaining an audit trail. These platforms enable secure AI adoption by provisioning access, monitoring AI activity, preventing data exposure, and collecting evidence for audits. SCIM-based access controls allow for automatic updates of AI tool access based on employee roles, while Data Loss Prevention (DLP) and observability layers ensure sensitive data is protected. Effective AI governance platforms offer flexible implementation, allowing companies to adapt quickly to new AI tools and maintain consistent policy enforcement. Vendor support is vital due to the complexity of implementation, involving mapping identity attributes, defining DLP rules, and configuring connectors. Leading AI governance platforms, such as Merge Agent Handler for Employees, Runlayer, and MintMCP, each offer unique strengths, like broad connector coverage, strong threat detection, and centralized MCP traffic management, although they may vary in scope, ease of use, and support.

Merge has introduced the Embedded Routing Stack within its Gateway, enabling customers to manage their own AI model routing, budgets, usage data, and policies through a single API. This feature allows users to bring their own LLM keys, set routing policies, and enforce budget limits, giving them full control over their AI models and optimizing token spend. As open-source models become more competitive, enterprises increasingly seek to self-host and choose models that align with their specific needs for cost, security, and scalability. The Embedded Routing Stack supports this demand by offering a customizable setup that integrates seamlessly with Merge’s platform, allowing businesses to maintain their model strategies without overhauling their backend systems.

Merge's platform offers an efficient way to connect the Gamma API with Codex through the use of the Merge Agent Handler and CLI, allowing for seamless integration of presentation automation tasks. By authenticating and managing API keys centrally, Merge ensures that Codex can access real-time Gamma data, such as presentation structures and themes, to produce accurate and reliable code. This integration is particularly beneficial for teams handling multiple workspaces or tools, as it provides scoped access and audit logging for secure and accountable operations. The platform supports provisioning and access control via SCIM, enabling IT teams to manage employee access and ensure compliance with security policies. Furthermore, Merge's solution extends to managing customer integrations, highlighting its role as more than just a Unified API product.

GLM-5.2 and Claude Sonnet 5 are two prominent coding models evaluated on their performance and cost-effectiveness, with GLM-5.2 known for its affordability and detailed output, while Claude Sonnet 5 is recognized for its speed and reliability, especially in mobile responsiveness. Released in June 2026, both models utilize a 1,000,000-token context window, but they differ significantly in pricing, with GLM-5.2 offering lower rates compared to Claude Sonnet 5's premium pricing. Evaluations using DesignArena's Fullstack App Quality rubric indicate Claude Sonnet 5 outperforms GLM-5.2 in areas such as UI quality and error handling, although GLM-5.2 is cited for producing more specific and detailed content at a lower cost. Despite this, Claude Sonnet 5 was quicker and maintained a higher level of responsiveness in a practical test scenario. Merge's Gateway service allows users to optimize their AI model selection based on various factors like budget, speed, and specific project requirements, offering flexibility and control over the routing of AI model requests.

In a detailed comparison between Anthropic's Claude Fable 5 and OpenAI's GPT-5.6 Sol, the document explores the capabilities, pricing, and performance of these two flagship AI models. Claude Fable 5, released in June 2026, is designed for complex, long-running tasks with a 1,000,000-token context window, priced at $10 per million input tokens and $50 per million output tokens. Conversely, GPT-5.6 Sol, released a month later, offers a slightly larger 1,050,000-token context window at a lower cost of $3.75 per million input tokens and $22.5 per million output tokens. Both models were tested on a coding task to build a homepage for a fictional company, revealing that while Claude Fable 5 completed the task faster and with fewer tokens, it was more expensive, whereas GPT-5.6 Sol provided a richer product story at a lower cost. The text emphasizes that Merge Gateway can be used to route tasks to the most suitable model, based on specific needs, such as cost and performance requirements.

Merge offers a platform that simplifies the integration of Supabase with Cursor by using the Merge Agent Handler and its Supabase MCP server. This setup allows developers to access real-time database structures, such as column types, constraints, and row-level security policies, directly within their development environment, eliminating the need to switch between the editor and external dashboards. The Merge CLI facilitates this connection, ensuring secure handling of Supabase credentials and enabling developers to perform various tasks like inspecting table schemas, verifying authorization logic, and managing storage objects efficiently. For organizations, Merge provides tools for provisioning, securing, and governing employee access to Supabase, ensuring that operations are controlled and logged. Additionally, Merge's integration platform supports over 160 connectors, enhancing the ability to manage customer integrations effectively.

Merge offers a robust integration platform that accelerates the transition of AI projects from proof of concept to production by connecting Codex with Oracle HCM through its Merge Agent Handler. This integration allows Codex to access real-time Oracle HCM data, ensuring accurate workforce automation code, syncs, and reporting tools by providing direct access to tenant-specific data structures, flexfields, and effective-dated histories. The setup involves installing and authenticating the Merge CLI, linking it to the Oracle HCM API, and configuring Codex to utilize this connection, which enhances security and operational efficiency by managing OAuth credentials and domain resolution centrally. With Merge, organizations can control and govern employee access to AI tools, ensuring compliance and data security while enabling the development of reliable automation processes.

Merge's blog post compares two AI models, Claude Sonnet 5 by Anthropic and Grok 4.5 by xAI, highlighting their performance in coding tasks. Claude Sonnet 5 offers a larger context window and is suited for quality-first, complex refactoring tasks, while Grok 4.5, with its faster and more cost-effective capabilities, excels in speed-sensitive and high-volume code generation tasks. In an experiment building a marketing homepage for a fictional company, Grok 4.5 outperformed Sonnet 5 in speed, cost, and rendering reliability, although both models delivered similar visual designs and copy. Despite Sonnet 5's superior navigation features, the failure to render an animated element marked a crucial disadvantage, making Grok 4.5 the better performer overall in this particular test. The blog underscores the flexibility of using Merge Gateway to route coding tasks to the most suitable model based on specific requirements like cost, speed, and task complexity.

Merge's Cookie Policy outlines the use of cookies to enhance user experience, requiring user consent as described in their Privacy Policy. Thousands of companies utilize Merge to expedite AI deployment from proof of concept to production. The text discusses the integration of UKG Pro with Codex via Merge Agent Handler, emphasizing how this connection allows Codex to access real-time data from UKG Pro, thereby improving the accuracy of workforce automation tasks. It details the setup process, including installing the Merge CLI and authenticating with UKG Pro, to facilitate seamless data retrieval for tasks such as syncs, reporting tools, and onboarding automations. The benefits of using Merge Agent Handler over a self-hosted UKG Pro MCP server are highlighted, noting centralized management, scoped access, and comprehensive audit logging as key advantages. Additionally, the integration supports IT and security teams in controlling data access and ensuring compliance through features like SCIM provisioning and policy enforcement.

The text discusses the role of MCP (Model Context Protocol) governance platforms in providing secure AI access for employees while offering IT departments the necessary security controls and visibility into AI usage. These platforms, such as Agent Handler for Employees, Runlayer, and MintMCP, facilitate the secure integration of AI tools like Claude, ChatGPT, and Cursor into business workflows by managing permissions, monitoring AI interactions, and ensuring compliance through audit logs. Key considerations for evaluating MCP governance platforms include centralized user experience, SCIM-based provisioning, auditability, observability, and comprehensive connector coverage. The text highlights the benefits and limitations of different platforms, emphasizing the need to assess their suitability based on features, security, and scalability to ensure a successful deployment within organizations.

The comparison between Kimi K2.6 and Claude Sonnet 4.6 highlights their strengths and weaknesses in handling coding tasks, particularly in terms of cost, speed, and output. Kimi K2.6, developed by Moonshot AI, is a cost-effective, high-context model with a 262,144-token context window and lower pricing, suitable for cost-sensitive and high-volume tasks such as bulk code generation and test scaffolding. In contrast, Claude Sonnet 4.6 from Anthropic boasts a 1,000,000-token context window and excels in quality and latency-sensitive tasks like complex refactors or debugging due to its faster response times and higher Coding Agent Index score. While Kimi offers a cheaper solution with slower task completion, Claude Sonnet provides a more polished output and faster execution. Merge Gateway facilitates the optimal use of these models by allowing users to route requests based on specific requirements, such as cost or latency, thus ensuring tasks are matched with the most suitable model.

The text compares two AI models, GPT-5.5 by OpenAI and DeepSeek V4 Pro by DeepSeek, highlighting their strengths and weaknesses based on coding performance and cost-efficiency. GPT-5.5 is a closed model known for its superior coding abilities and faster outputs, offering high-quality results at a premium price. In contrast, DeepSeek V4 Pro, with its open-weight architecture, is more cost-effective and allows self-hosting, making it ideal for high-volume and cost-sensitive coding tasks. The analysis includes a practical test where both models generated marketing homepage content, demonstrating GPT-5.5's superior quality but higher cost compared to DeepSeek's more affordable, albeit less polished, output. The Merge Gateway platform is presented as a solution for routing coding tasks to the most appropriate model based on specific requirements, providing flexibility to switch models as conditions change.

Merge's platform facilitates the integration of AI tools like Codex with Notion by using the Merge Agent Handler, which connects Codex directly to Notion's API through the Merge CLI. This process allows Codex to access Notion's data directly, reducing errors caused by paraphrasing and assumptions, and enabling more accurate code generation based on actual specifications and requirements. The Merge Agent Handler centralizes authentication, manages OAuth credentials, and provides a control layer for defining and enforcing operational boundaries, ensuring both security and observability of the integration. This setup is particularly beneficial for organizations looking to provision, secure, and govern employee access to AI tools, allowing for centralized control over content access while maintaining a comprehensive audit trail of interactions.

Merge's platform offers a sophisticated approach to managing language model (LLM) requests by combining both proxy and router functionalities. While proxies serve as intermediaries between applications and model providers, ensuring unified access, logging, and resilience, routers make decisions on which models should handle specific requests based on factors like cost, latency, and quality. The article emphasizes the importance of distinguishing between these two layers, as they address different challenges in AI infrastructure. In practice, most tools integrate both functions under what is known as an LLM gateway, allowing users to efficiently route mixed-difficulty traffic and manage high LLM expenditures. Merge's Gateway exemplifies this by providing a unified endpoint that not only facilitates seamless transport but also enhances decision-making for routing requests across multiple models. For organizations, the order of implementation is crucial: proxies should be established first to gain visibility, which then informs the development of effective routing rules. Ultimately, Merge is highlighted as more than just a Unified API product, but as an integration platform capable of managing customer integrations and optimizing AI operations.

Merge leverages its Gong connector and other tools to enhance its go-to-market (GTM) strategies by analyzing customer feedback and improving communication within the organization. By implementing Claude skills, the company automates the extraction of actionable insights from sales calls, which are then shared with teams via Slack, allowing them to address product strengths and weaknesses, and capitalize on new features to regain closed-lost deals. The content team uses a similar approach to prioritize content creation by analyzing call transcripts and search volumes, helping allocate resources effectively. Jon Gitlin, the Senior Content Marketing Manager, emphasizes that customer feedback is invaluable for GTM success, and Merge continues to explore AI-powered automations to further refine its operations and share insights.