{"id":6025,"date":"2026-07-15T19:52:10","date_gmt":"2026-07-15T19:52:10","guid":{"rendered":"https:\/\/www.gosearch.ai\/blog\/?p=6025"},"modified":"2026-07-15T19:52:11","modified_gmt":"2026-07-15T19:52:11","slug":"whats-new-in-claude-sonnet-5-powering-enterprise-ai-agents","status":"publish","type":"post","link":"https:\/\/www.gosearch.ai\/blog\/whats-new-in-claude-sonnet-5-powering-enterprise-ai-agents\/","title":{"rendered":"What’s New in Claude Sonnet 5: Powering Enterprise AI, Agents, and Workflows in GoSearch"},"content":{"rendered":"\n

GoSearch is an enterprise AI platform that connects to the tools and systems your company already uses, letting teams search across them, chat with an AI assistant that has real context, and build no-code agents and automated workflows without writing code. Anthropic released Claude Sonnet 5 on June 30, 2026, calling it the most agentic Sonnet model to date, and it’s now available across GoSearch’s no-code agent builder, workflow automation, and GoAI, GoSearch’s conversational AI chat.<\/p>\n\n\n\n

This post covers what actually changed in Sonnet 5, how it compares to its predecessor, and what it means for teams building agents and automating work inside GoSearch.<\/p>\n\n\n\n

Quick answer:<\/strong> Claude Sonnet 5 is Anthropic’s latest mid-tier model, built to bring flagship-level agentic performance, planning, tool use, and multi-step task execution, into a faster, lower-cost model class. It scores 80.5% on the Terminal-bench 2.1 agentic coding benchmark<\/a>, up from 67% for Sonnet 4.6, and Anthropic reports it approaches Claude Opus 4.8 on several complex tasks at a fraction of the cost. For enterprise teams, that means agents that used to require a flagship-tier model can often run on Sonnet 5 instead, with adjustable effort levels to balance cost and performance per task. In GoSearch, Sonnet 5 is available alongside models from OpenAI and Google as part of the platform’s multi-LLM approach, all with Zero Data Retention.<\/p>\n\n\n\n

<\/div>\n\n\n\n

What’s new in Claude Sonnet 5?<\/h2>\n\n\n\n

Anthropic built Sonnet 5 around a single theme: agentic execution. Where earlier Sonnet models were strong at answering questions and writing code, Sonnet 5 is designed to plan a task, use tools like browsers and terminals, execute multiple steps in sequence, and keep working through longer task chains without losing track of the goal.<\/p>\n\n\n\n

A few specific changes stand out. Anthropic reports<\/a> that Sonnet 5 has a lower overall rate of problematic behaviors compared to Sonnet 4.6 and is safer to deploy in agentic settings, while independent coverage from TechCrunch<\/a> noted that testers observed the model reviewing and validating its own work without being prompted to, a meaningful shift for agents that run with minimal supervision. The model also introduces adjustable effort levels, letting teams dial performance up for harder tasks or down for routine ones, rather than choosing a single fixed model tier for everything.<\/p>\n\n\n\n

Sonnet 5 also ships with an updated tokenizer, meaning the same input may now map to somewhat more tokens than it did under Sonnet 4.6, worth factoring in if you’re estimating usage against a token-based plan. However, GoSearch does not price based on token usage. <\/p>\n\n\n\n

<\/div>\n\n\n\n

Claude Sonnet 5 vs. Sonnet 4.6<\/h2>\n\n\n\n
<\/div>\n\n\n\n
<\/th>Claude Sonnet 4.6<\/strong><\/th>Claude Sonnet 5<\/strong><\/th><\/tr><\/thead>
Agentic coding (Terminal-bench 2.1)<\/strong><\/td>67%<\/td>80.5%<\/td><\/tr>
Best for<\/strong><\/td>Standard coding and chat tasks<\/td>Multi-step agentic work, tool use, and complex coding<\/td><\/tr>
Effort control<\/strong><\/td>Fixed performance tier<\/td>Adjustable effort levels to balance cost and performance<\/td><\/tr>
Self-verification<\/strong><\/td>Limited<\/td>Checks its own output on complex tasks without being asked<\/td><\/tr>
Relative capability<\/strong><\/td>Strong for its tier<\/td>Approaches Opus-class performance on several complex tasks<\/td><\/tr><\/tbody><\/table>
Claude Sonnet 4.6 vs. Sonnet 5 for enterprise AI<\/figcaption><\/figure>\n\n\n\n
<\/div>\n\n\n\n

The practical takeaway: Sonnet 5 doesn’t just do the same tasks a little better, it closes much of the gap that used to require stepping up to a flagship-tier model. For many enterprise use cases, that changes the default choice from “start with the flagship model and downgrade if it’s too expensive” to “start with Sonnet 5 and step up only when a task genuinely needs it.”<\/p>\n\n\n\n

<\/div>\n\n\n\n

How does the model power enterprise AI agents and workflows?<\/h2>\n\n\n\n

The clearest signal of what Sonnet 5 is built for comes from how early enterprise users describe it. Anthropic’s case studies point to insurance workflows, where computer-use agents built on Sonnet 5 reliably handle submission intake and claims processing on existing systems, and to automation platforms like Zapier, where testers found the model could complete multi-step business tasks, like updating a CRM and sending a follow-up communication, end to end without stalling partway through the way earlier models sometimes did.<\/p>\n\n\n\n

That pattern matters for the kind of work GoSearch customers build agents around: tasks that touch multiple systems, require several steps done in the right order, and need to actually finish rather than hand back a partial result.<\/p>\n\n\n\n

Benefits by team:<\/strong><\/p>\n\n\n\n