Claude 3.5 Sonnet is no longer a current Claude model. Anthropic retired both Claude 3.5 Sonnet API versions on 28 October 2025.
That does not make the model unimportant. Claude 3.5 Sonnet was a major release in 2024 and helped set expectations for strong writing, coding, vision, and long-context work at a mid-tier price. Many older articles, prompts, and software integrations still mention it.
If you are choosing a model for new work in 2026, use a current Claude model. If you are researching Claude 3.5 Sonnet because an old workflow depends on it, plan and test a migration.
This guide explains what Claude 3.5 Sonnet could do, why it was retired, and how to choose a replacement.
Claude 3.5 Sonnet Status in 2026
Anthropic’s model deprecation history lists both Claude 3.5 Sonnet versions as retired:
claude-3-5-sonnet-20240620claude-3-5-sonnet-20241022
Requests to retired models fail on Anthropic-operated platforms. Partner platforms can use different schedules, so check the provider that hosts your workload rather than assuming the same date applies everywhere.
Anthropic currently recommends moving older Sonnet workloads to a current Sonnet model and testing the replacement against your own tasks.
What Made Claude 3.5 Sonnet Important?
Anthropic launched Claude 3.5 Sonnet in June 2024. Its original announcement positioned it as a step forward in graduate-level reasoning, coding, visual reasoning, and speed.
Several capabilities made it widely used.
A 200,000-token context window
Claude 3.5 Sonnet could process long documents and conversations in one request. That made it useful for:
- Reviewing reports and contracts
- Summarising research
- Working across larger code samples
- Comparing multiple documents
- Extracting information from lengthy source material
Context size did not guarantee a correct answer. Teams still needed good document structure, clear instructions, and checks against the source.
Strong writing and instruction following
The model became popular for drafting, rewriting, summarising, and matching a requested format or tone. It often produced natural prose without requiring a long prompt.
That strength still matters when evaluating a replacement. A benchmark score alone does not tell you whether a model will follow your house style, use the right evidence, or return a result in the format your workflow needs.
Coding and tool use
Claude 3.5 Sonnet was widely used for generating, explaining, reviewing, and debugging code. The updated October 2024 version also arrived with Anthropic’s early computer-use capability.
Current Claude models go further in planning and tool use, but the same testing rule applies: evaluate the model on your repository, tools, and acceptance criteria.
Vision
Claude 3.5 Sonnet accepted text and image input. It could interpret screenshots, charts, diagrams, and scanned material when the visual content was clear.
For an operational workflow, vision is most useful when it ends in a structured action. Examples include extracting invoice fields for review, reading a form, or checking a dashboard screenshot before creating a report.
Claude 3.5 Sonnet vs Current Claude Models
Anthropic’s current models overview lists a newer group of models with different speed, capability, and cost profiles.
| Model | Best fit | Context window listed by Anthropic |
|---|---|---|
| Claude Haiku 4.5 | Fast, high-volume tasks where lower latency matters | 200K |
| Claude Sonnet 5 | A balance of speed and capability for general production work | 1M |
| Claude Opus 4.8 | Complex agentic coding and enterprise work | 1M |
| Claude Fable 5 | Long-running work that needs Anthropic’s highest widely available capability | 1M |
Capabilities, availability, and prices can change. Check Anthropic’s current documentation before making a production choice.
The practical replacement for most Sonnet workloads
For most teams moving from Claude 3.5 Sonnet, Claude Sonnet 5 is the first model to test. Anthropic describes it as the best combination of speed and intelligence and announced it as a more agentic Sonnet model in June 2026.
Do not switch the model name and assume the output will stay the same. Newer models can interpret prompts differently, use different tokenisation, and apply different defaults.
When to consider Haiku
Use a current Haiku model when speed and cost matter more than maximum reasoning depth. Good candidates include:
- Classifying routine requests
- Extracting a small set of fields
- Reformatting clean text
- Routing work to the right queue
- Drafting a first-pass summary
Test difficult edge cases before using a smaller model for decisions that affect money, access, compliance, or customers.
When to consider Opus or Fable
Use a higher-capability model when the task needs long planning, complex reasoning, extensive tool use, or difficult coding.
A larger model is not automatically better for every step. A workflow can use a fast model for classification and a stronger one only when the task becomes complex. That can improve both response time and cost.
How to Migrate From Claude 3.5 Sonnet
Treat a model migration like a product change, not a configuration update.
1. Find every dependency
Search for the retired model ID in:
- Application code
- Environment variables
- Automation tools
- Prompt management systems
- Tests
- Documentation
- Saved templates
Check partner platforms separately because their model names and retirement dates can differ.
2. Create a representative test set
Use real examples from the workflow, with sensitive data removed where needed. Include:
- Common requests
- Long inputs
- Ambiguous instructions
- Missing information
- Invalid formats
- Safety boundaries
- Tasks that require tools
- Examples the old model handled poorly
Define the expected result before testing the replacement.
3. Compare outcomes that matter
Measure more than whether the response looks good.
Track:
- Accuracy
- Source use
- Required fields
- Format compliance
- Tool-call success
- Latency
- Token usage
- Human edit rate
- Escalation rate
A newer model can be more capable while still requiring prompt or workflow changes.
4. Update prompts and limits
Review system instructions, output limits, retry rules, and structured-output settings. Remove prompt workarounds that were only needed for the old model.
5. Run a controlled rollout
Send a small share of work to the replacement, compare results, and keep a rollback path until the new version is stable.
Choosing a Model for Internal Operations
The best model is the one that completes the whole task reliably inside your workflow.
Start with the job:
- Document extraction: Can it return the required fields and flag uncertainty?
- Knowledge assistant: Can it answer from approved sources and say when evidence is missing?
- Approval workflow: Can it prepare a recommendation without making the final decision?
- CRM assistant: Can it summarise account history and draft the next action without changing records incorrectly?
- Operations report: Can it use the right data, explain anomalies, and produce a consistent format?
Then choose the smallest model that passes your test set. Keep a person in control of high-impact decisions.
Use Claude in an Appaca Workflow
Appaca is an AI workspace for operators. It combines internal apps, a shared database, team knowledge, integrations, and AI coworkers.
Instead of putting a model behind a blank chat box, you can give it a defined place in a process. For example:
- Read a new invoice and prepare fields for finance review
- Answer an SOP question from approved team knowledge
- Draft a follow-up from CRM activity
- Turn onboarding information into a checklist
- Prepare a weekly operations summary from workspace data
The model handles the language task. The app provides the records, permissions, status, and human review around it.
Build an AI workflow in Appaca and test it with the real examples your team already handles.