Claude Opus 4 and Sonnet 4 launch amid AI boom, as Anthropic’s annualised revenue hits US$2bn
Anthropic has launched its most powerful AI models to date—Claude Opus 4 and Claude Sonnet 4—positioning the company at the forefront of AI development with claims that Opus 4 is now the world’s best coding model.
Backed by Amazon and founded by former OpenAI executives, Anthropic announced the Claude 4 family on Thursday. The models feature substantial improvements in coding, autonomous task execution, and reasoning, with Opus 4 reportedly capable of sustaining performance for up to seven hours—effectively simulating a full corporate workday.
“This is not just an incremental step,” said chief science officer Jared Kaplan. “These models are much, much stronger as agents and as coders.”
The new frontier in agentic AI
Claude Opus 4 and Sonnet 4 are designed with hybrid reasoning systems—able to switch between near-instant responses and extended, tool-assisted thinking. Both models can independently search the web, extract and store key information, and use multiple tools in parallel, which Anthropic says enables more coherent long-running tasks. When given access to local files, they maintain continuity over time, building what Anthropic calls “tacit knowledge.”
The release includes a notable performance claim: Opus 4 leads the SWE-bench coding benchmark at 72.5%, surpassing OpenAI’s GPT-4.1 and Google’s Gemini 2.5 Pro. With support for memory files, extended context windows, and custom tool use, Opus 4 is being adopted across enterprise applications—from writing full-stack software to handling complex pull request reviews and automated debugging.
Sonnet 4, while less powerful, builds on the prior Sonnet 3.7 model with improved instruction following and problem-solving. GitHub has already integrated it into its upcoming Copilot agent, praising its handling of agentic tasks and code reviews.
Claude Code and developer tools expand
The launch coincides with the general release of Claude Code, Anthropic’s developer assistant, which now integrates directly into IDEs like VS Code and JetBrains. Developers can invoke Claude to review pull requests, address CI errors, and propose code edits inline.
The company has also expanded its developer API offerings, adding support for code execution tools, GitHub integrations, file management, and memory caching. Pricing for Claude Opus 4 is set at US$15 per million input tokens and US$75 per million output tokens, while Sonnet 4 remains significantly cheaper at US$3/$15.
Revenue growth and safety risks
Anthropic’s revenue has surged alongside the model upgrades, doubling from US$1bn to US$2bn annualised in Q1 2025. The number of customers spending over US$100,000 annually has increased eightfold, and the company recently secured a US$2.5bn revolving credit line to support continued development.
But the launch also comes with serious safety caveats. Claude Opus 4 is the first model to trigger Anthropic’s ASL-3 (AI Safety Level 3) protocols, after internal testing showed it could potentially assist users with basic technical backgrounds in constructing chemical or biological weapons.
To mitigate these risks, Anthropic has implemented more than 100 safety controls, including real-time monitoring, restricted egress bandwidth, two-person authorisation for certain operations, and a bug bounty program offering up to US$25,000.
Shifting competitive landscape
The Claude 4 release places Anthropic in direct competition with OpenAI, Google, and Meta in the race to develop general-purpose AI agents. While OpenAI’s models still lead in general reasoning and multimodal capabilities, Anthropic is staking its lead in sustained, autonomous coding and planning tasks.
Anthropic’s transparency around safety concerns may also set it apart in an increasingly scrutinised field. “This is a large step toward the virtual collaborator,” said the company in its announcement. “These models maintain full context, sustain focus, and can drive transformational impact.”
Claude Opus 4 and Sonnet 4 are now available via Anthropic’s API, Amazon Bedrock, and Google Cloud Vertex AI. Sonnet 4 is also accessible to free users of Claude, while Opus 4 is included in the Claude Pro, Max, Team, and Enterprise tiers.
