Anthropic announced Claude Opus 4 and Claude Sonnet 4 on May 22, 2025. The launch brought two models, hybrid reasoning modes, expanded tool capabilities and new developer features—not one model that was simply better at everything. This is a look back at what changed at launch; Anthropic has since announced Sonnet 4.6 and Opus 4.8.
What Claude 4 meant at launch
“Claude 4” referred to two models: Opus 4 and Sonnet 4. Anthropic positioned Opus as its most capable option for difficult coding and sustained, complex tasks, while Sonnet was designed to balance capability, efficiency and steerability. Those descriptions are Anthropic’s product positioning, not an independent comparison across AI providers. Anthropic’s May 2025 announcement gives the launch details.
![Claude AI for Beginners Bible: [5 in 1] The Ultimate Guide to Automate Your Work, Save Hours Every Week, and Use AI for Real-World Results](https://m.media-amazon.com/images/I/415+fSJacsL._SL160_.jpg)
Here are ten notable changes and capabilities described at the time.
The 10 changes in the Claude 4 launch
1. Two models for different workloads
Opus 4 and Sonnet 4 launched together. The distinction was mainly about the work and trade-offs: Opus was aimed at more demanding, extended tasks, while Sonnet was the more efficient choice for capable everyday use. Neither was presented as universally best for every job.
2. A choice between fast responses and extended thinking
Both models offered a standard mode for near-instant responses and an extended-thinking mode for deeper reasoning. In practice, this gave developers a way to choose between speed and additional reasoning effort rather than treating every request identically. See the launch announcement and Claude 4 System Card.
3. Tool use while extended thinking was in progress
Anthropic announced tool use during extended thinking in beta. That mattered for agent-style work: a model could reason through a task and use tools as part of that process. The beta qualification applies to the launch announcement; it should not be read as a statement about present availability.
4. Parallel tool use
Anthropic said both models could use tools in parallel. For applications that expose suitable tools, parallel calls can let a model gather or act on multiple pieces of information without handling every operation strictly one at a time. The capability depends on the application and its tool setup.
5. More precise instruction following
Anthropic highlighted improved responsiveness to user instructions. This was a claimed launch improvement in how the models followed directions, not a guarantee that they would always interpret an ambiguous or conflicting prompt correctly.
Crashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minuteWindows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstall6. Memory through developer-provided local files
Anthropic described models extracting and saving important facts for continuity when developers gave them access to local files. That is an application-level capability: Claude does not thereby gain independent access to a user’s computer. The software using the model must provide file access and determine what may be read or written.
7. Claude Code moved out of research preview
The launch post said Claude Code had graduated from research preview. Anthropic also cited background tasks through GitHub Actions and native integrations for VS Code and JetBrains IDEs. These describe the launch-era offering; the announcement is not a guide to current Claude Code features or availability.
8. Code execution for API agents
In a companion API announcement, Anthropic introduced a code execution tool for agent applications. It was intended to let an application give Claude a way to run code as part of a broader workflow. Developers still control the environment and permissions in which code runs. See Anthropic’s API capabilities announcement.
9. Remote MCP, a Files API and longer prompt caching
The same API post grouped remote Model Context Protocol (MCP), a Files API and prompt caching of up to one hour among its new capabilities. Those are launch-era details, not confirmation of current limits or beta status. Developers should consult current Anthropic documentation before designing around a particular API feature or cache window.
Do these 3 things before closing this tab:
1Scan for outdated or missing drivers - takes under a minute2Clear out junk files and repair common Windows errors3Fix the driver behind crashes, sound loss and screen glitches10. Multiple launch access routes
At launch, Anthropic said both models were available through its API, Amazon Bedrock and Google Cloud Vertex AI; Sonnet 4 was also available to free Claude users. This records the May 2025 announcement, not a guarantee about current access, regions or plans.
Opus 4 or Sonnet 4: how the launch-era choice differed
The practical choice depended on task demands, desired reasoning behavior and API budget. The figures below are Anthropic’s reported launch results and launch API rates, not current pricing or an independent head-to-head test.
| Comparison | Claude Opus 4 | Claude Sonnet 4 |
|---|---|---|
| Launch positioning | Advanced coding, difficult problem-solving and sustained complex tasks. | Capability balanced with efficiency and steerability. |
| Reported coding benchmarks | 72.5% on SWE-bench Verified and 43.2% on Terminal-bench, as reported by Anthropic in May 2025. Anthropic said these results used no extended thinking. | 72.7% on SWE-bench, as reported by Anthropic in May 2025. Anthropic said this result used no extended thinking. |
| Reported agent-task comparison | Not stated for Opus 4 in the cited launch comparison. | Anthropic said Sonnet 4 was 65% less likely than Sonnet 3.7 to use shortcuts or loopholes on agentic tasks especially susceptible to them. This is Anthropic’s own evaluation. |
| Reasoning and tools | Standard and extended-thinking modes; Anthropic also announced tool use during extended thinking in beta and parallel tool use. | Standard and extended-thinking modes; Anthropic also announced tool use during extended thinking in beta and parallel tool use. |
| Launch API price per million tokens | $15 per million input tokens and $75 per million output tokens, as stated by Anthropic at launch in May 2025. | $3 per million input tokens and $15 per million output tokens, as stated by Anthropic at launch in May 2025. |
Anthropic said the benchmark figures in its announcement were the highest results achieved with or without extended thinking, and specified that the SWE-bench Verified and Terminal-bench results shown used no extended thinking. Other results in the post used extended thinking of up to 64K tokens. These figures are useful context for the company’s claims, but benchmarks do not establish how a model will perform on every codebase or workflow. The launch announcement is the source for the benchmark results and methodology notes.
Anthropic also reported that Rakuten validated Opus 4 on an open-source refactor that it ran independently for seven hours. That is a customer example reported in Anthropic’s launch post, not a general guarantee that Opus can work unattended for seven hours.
Free tools Windows power users keep installed
One-click scans. No signup required.
Best Value
What the Claude 4 System Card says about training and safety
Anthropic’s May 2025 System Card says Opus 4 was released under its AI Safety Level 3 Standard and Sonnet 4 under its AI Safety Level 2 Standard. The card describes a proprietary training mix that included public internet information available through March 2025, non-public third-party data, data-labeling services and paid contractors, opted-in Claude-user data, and internally generated material. These are company disclosures, not independent audits. The Claude 4 System Card also says both models were trained with a focus on being helpful, honest and harmless; that statement describes intent, not a guarantee of behavior.
How to read the launch prices and availability now
The API prices in the comparison table are historical May 2025 launch rates. Do not use them as current rates: check Anthropic’s current official pricing before estimating a project’s cost. Similarly, the listed providers and free-plan access describe what Anthropic announced at launch and may not match current availability.
Claude 4 is no longer the newest generation described in Anthropic’s announcements. Anthropic announced Sonnet 4.6 on February 17, 2026, describing a 1-million-token context window in beta and naming it the default for Free and Pro plans at that time. It announced Opus 4.8 on May 28, 2026. Those later releases do not change what the original Claude 4 models offered in May 2025, and their features should not be attributed retroactively to Opus 4 or Sonnet 4. See the official announcements for Sonnet 4.6 and Opus 4.8.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
The Tool Desk
Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →
