Feature
Anthropic's Claude 3.7 Sonnet landed in late February as the first "hybrid reasoning" model on the market. Hybrid reasoning means users can choose between "normal" reasoning — the rapid responses we've grown accustomed to — "extended," more involved responses whose step-by-step "thinking" process is visible to the end user.
Claude 3.7 Sonnet is available to everyone, however only paid users have access to the reasoning option. The price carries over from previous models, with one million input tokens costing $3 and one million output tokens costing $15. Claude 3.7 Sonnet can also be accessed across platforms like Amazon Bedrock, Google Cloud's Vertex AI and Anthropic's API. The release comes only four months after the release of Claude 3.5 Sonnet.
What's Inside the Claude 3.7 Sonnet Release
Claude 3.7 Sonnet builds upon its predecessor, Claude 3.5 Sonnet, offering substantial improvements in reasoning, coding and task execution. What really makes it stand out from the others though, is:
1. Hybrid Reasoning Modes
As mentioned before, the model operates in two modes — standard and extended thinking. In standard mode, it provides quick responses optimized for tasks like knowledge retrieval and automation. In extended thinking mode, it breaks down problems into manageable steps, visually displays its reasoning process, and delivers highly accurate solutions for complex challenges such as coding, physics problems and strategic analysis.
While competitors offer the same functionality, it is split across different models (think the difference between OpenAI's ChatGPT 4o model and its ChatGPT o1 model), rather than combining it in a single model. OpenAI also has a hybrid reasoning model in the works, according to Sam Altman.
OPENAI ROADMAP UPDATE FOR GPT-4.5 and GPT-5:
— Sam Altman (@sama) February 12, 2025
We want to do a better job of sharing our intended roadmap, and a much better job simplifying our product offerings.
We want AI to “just work” for you; we realize how complicated our model and product offerings have gotten.
We hate…
2. Enhanced Coding Capabilities
Claude 3.7 excels in software engineering tasks, scoring an industry-leading accuracy of 70.3% on SWE-bench Verified when using structured prompts. It supports end-to-end software development processes, including debugging, refactoring and creating production-ready code across multiple languages.
3. Expanded Output Capacity
The model can generate outputs up to 128K tokens long (64K tokens generally available), which allows for detailed analyses, comprehensive outlines or extensive content generation.
4. Agentic Tool Use
Its improved performance in tasks requiring interaction with digital interfaces — such as clicking buttons or typing text — makes it suitable for automating workflows and testing applications. According to Anthropic, Claude 3.7 Sonnet is optimized for diverse use cases:
- Software development: From planning to debugging and maintenance, the model supports complex coding tasks with reduced error rates and improved precision.
- Customer-facing AI agents: Its advanced reasoning capabilities enhance customer support workflows by improving instruction-following and tool selection.
- Strategic analysis: Extended thinking mode allows businesses to tackle intricate problems requiring deep reflection and multi-step planning.
Anthropic has positioned Claude 3.7 as its most intelligent model to date, which promises significant improvements in coding, front-end development and enterprise workflow automation.
Adding to its appeal, Anthropic introduced Claude Code in limited research preview, a command-line tool that allows developers to offload substantial engineering tasks to AI agents directly from their terminal.
Transparency and Collaborative Development Stand Out
The hybrid reasoning capabilities are one of the most noteworthy elements in the new release, Andrew Reece, chief AI scientist at BetterUp, told Reworked.
“The hybrid reasoning approach might best be understood as giving you options in how your AI partner thinks — sometimes you need quick, efficient responses, and other times you want more deliberate analysis,” he said.
The visibility into the thinking process, which creates a level of transparency that helps bridge the gap between human and machine cognition, is potentially transformative, Reece added.
The value proposition for enterprise users is the flexibility to match the AI's thinking mode to the task complexity, potentially allowing people to redirect their cognitive bandwidth toward refining strategic vision rather than managing execution details.
Claude Sonnet 3.7's improvements in coding and front-end development offer what is effectively an evolution from prediction-based code assistance to something more closely resembling collaborative development, Reece continued.
“The interesting shift isn't necessarily in code quality — although Sonnet 3.7’s coding chops are head-and-shoulders above the competition — but in how this new class of models reduces the cognitive load of translating high-level ideas into implementation details,” he said.
This change moves users towards a best-case scenario, where developers are able to spend more time on things like visioning, planning and design — areas where human judgment adds unique value. People call it “vibe coding,” but Reece predicts in time we won’t even think of it as “coding” at all, it will just be creating.
Technical Barriers Might Prevent Knowledge Workers From Fully Benefitting (for Now)
Coders and software engineers aren't the only people likely to benefit. Knowledge workers also can benefit from Claude 3.7 Sonnet's capabilities, Pmfm.ai founder Aditya Saxena told Reworked.
He cites the example of a writer who trains a custom Claude 3.7 Sonnet bot on their style to use as an assistant, a lawyer using it to draft briefs or a small business tapping it for market research, report generation or document analysis.
“However, it’s fair to say that its impact would be felt the most in the realm of software development,” he said, due to the complexities of using APIs and setting up AI agents.
So while Saxena sees the potential productivity gains for knowledge workers, he expects adoption will remain low due to the high technical barrier.
Conference
Sep
16
HR Tech Conference Las Vegas 2025
Register
Conference
Oct
27
Gartner HR Symposium/Xpo Orlando 2025
Register
He also reminds businesses that no matter how sophisticated a model is — including Claude 3.7 Sonnet — humans still need to be in the loop, particularly in critical processes like code deployment or review.
“AI models will incrementally get better at agentic reasoning and automating workflows. That is inevitable. But that shouldn’t translate into humans playing a less critical role by any means," Saxena said. “Humans have to ultimately take ownership of their work — whether it’s done by AI or not. All organizations using AI agents should know that LLMs aren’t infallible and all AI work should be scrutinized by a human."
Claude Code Is the Blueprint for What's to Come
TetraNoodle Technologies founder and CIO Manuj Aggarwal echoes Saxena's reminder, stating AI is just a tool, not a decision-maker. “I advise business leaders to let AI handle execution, but keep humans in charge of strategy and critical decisions. Set clear guardrails, monitor and audit the decisions made by AI, and use AI as a co-worker for your workforce. AI should assist, not replace humans. The companies that get this right will scale faster, work smarter and avoid costly mistakes."
Reasoning models, in their ability to peel back the curtain on their thought processes, will help enterprises get familiar with how generative AI “thinks,” which can help in building trust in results, said Reece.
While Sonnet 3.7 is an impressive model, it is just one in a cohort of models with these advanced capabilities, he continued. “I’m not sure it’s going to be a game-changer for knowledge work in the enterprise on its own. Claude Code is the blueprint, though, for where things are headed, in terms of a human-agent hybrid workforce as the dominant labor model for most enterprises,” he said.
Editor's Note: Read more about other recent developments in AI:
- Salesforce Brings an End to the 'Work of Work' — With Agentforce 2.0, Salesforce envisions a future of work where thousands of agents pick up the slack in Slack and beyond.
- GenAI Comes for Project Management: How Asana, ClickUp, Monday and Wrike Are Evolving — A look at how four project management platform leaders have incorporated generative AI features to improve project management.
- Who Watches the AI? Why Agentic AI Needs Observability Platforms — Agentic AI has greater potential but also higher risks than traditional LLMs. Observability platforms play a part in making sure things don't go off the rails.