Claude Sonnet 5
Last reviewed
Sources
5 citations
Review status
Source-backed
Revision
v1 · 1,280 words
Claude Sonnet 5 is a large language model developed by Anthropic, released on June 30, 2026 as the mid-tier, default member of the current Claude lineup. [1][4] It is the next generation of Anthropic's Sonnet class, positioned below the Claude Opus 4.8 and Claude Fable 5 flagships and above Claude Haiku 4.5, and Anthropic markets it as "the best combination of speed and intelligence." [1][3] Sonnet 5 launched at standard API pricing of $3 per million input tokens and $15 per million output tokens, with introductory pricing of $2 / $10 through August 31, 2026, and Anthropic describes its performance as "close to that of Opus 4.8, but at lower prices." [1][2][3]
What is Claude Sonnet 5?
Claude Sonnet 5 is a hosted, closed-weight, text-and-vision generative model and the direct successor to Claude Sonnet 4.6. [2][3] Anthropic calls it "the most agentic Sonnet model yet," built to plan, use tools such as browsers and terminals, and run autonomously at a level that previously required larger, more expensive models. [1] It is the default model for Free and Pro users on claude.ai and is available to Max, Team, and Enterprise customers, in Claude Code, and across the Claude API. [1][2]
Within Anthropic's tier structure, the "Sonnet" name denotes the balanced middle option between the smaller, faster Haiku models and the larger Opus and Fable models. Sonnet 5 shares the "5" generation label with Claude Fable 5 and the restricted Claude Mythos 5, both released June 9, 2026, while the Opus and Haiku tiers currently remain at Opus 4.8 and Haiku 4.5. [3] Fable 5 is Anthropic's most capable widely released model, and Mythos 5 is offered only through the invitation-only Project Glasswing. [3] The API model identifier is claude-sonnet-5, and the model carries a reliable knowledge cutoff of January 2026. [2][3]
| Specification | Claude Sonnet 5 |
|---|---|
| Developer | Anthropic |
| Released | June 30, 2026 |
| API model ID | claude-sonnet-5 |
| Model tier | Sonnet (mid-tier, default) |
| Predecessor | Claude Sonnet 4.6 |
| Context window | 1,000,000 tokens (default and maximum) |
| Max output | 128,000 tokens (up to 300,000 via Batch API beta) |
| Modalities | Text and image input, text output |
| Reasoning | Adaptive thinking, on by default |
| Knowledge cutoff | January 2026 |
| Standard price | $3 / $15 per million input / output tokens |
| Introductory price | $2 / $10 per million tokens, through Aug 31, 2026 |
How much does Claude Sonnet 5 cost?
Standard API pricing is $3 per million input tokens and $15 per million output tokens, unchanged from Claude Sonnet 4.6. [2] Through August 31, 2026, Anthropic offers introductory pricing of $2 per million input tokens and $10 per million output tokens. [1][2][3] Batch processing and prompt-caching discounts apply on top of these rates. [2]
One caveat affects real cost: Sonnet 5 ships with a new tokenizer that produces roughly 30% more tokens for the same text than Sonnet 4.6, so the per-token price is unchanged but the token count, and therefore the cost, of an equivalent request can be higher. [2] The independent evaluator Artificial Analysis noted that, because the model is verbose and reasons at length, a task run without the promotional pricing can cost more than the same task on Opus 4.8 despite Sonnet 5's lower per-token rate. [5] For comparison, Anthropic's other current models are priced at $10 / $50 (Fable 5), $5 / $25 (Opus 4.8), and $1 / $5 (Haiku 4.5) per million input / output tokens. [3]
What is Claude Sonnet 5 best at?
Anthropic reports that Sonnet 5's largest gains over Sonnet 4.6 are in coding and agentic tasks such as reasoning, tool use, and computer use. [1][2] On its official system card, Claude Sonnet 5 scored 85.2% on SWE-bench Verified, the widely cited real-world software-engineering benchmark, using adaptive thinking at maximum effort averaged over five trials. [4] It also posted strong results on agentic and computer-use evaluations. The table below shows selected system-card figures, with Sonnet 4.6 and OpenAI's GPT-5.5 as comparison points. [4]
| Benchmark | Claude Sonnet 5 | Claude Sonnet 4.6 | GPT-5.5 |
|---|---|---|---|
| SWE-bench Pro | 63.2% | 58.1% | 58.6% |
| Terminal-Bench 2.1 | 80.4% | 67.0% | 83.4% |
| OSWorld-Verified | 81.2% | 78.5% | 78.7% |
| BrowseComp (single agent) | 84.7% | 76.2% | 84.4% |
| Humanity's Last Exam (with tools) | 57.4% | 46.8% | 52.2% |
| Humanity's Last Exam (no tools) | 43.2% | 34.6% | 41.4% |
Anthropic's Terminal-Bench figure for GPT-5.5 uses the Codex CLI harness. [4] Beyond the table, Sonnet 5 scored 78.3% on SWE-bench Multilingual, 86.6% on multi-agent BrowseComp, and 1,618 Elo on GDPval-AA v2, an economically grounded task benchmark. [4] On the independent Artificial Analysis Intelligence Index, a composite of nine evaluations spanning reasoning, knowledge, math, and coding, Claude Sonnet 5 (adaptive reasoning, maximum effort) scored 53, ranking fifth among tracked models and improving six points over Sonnet 4.6. [5]
How does Sonnet 5 compare to Opus 4.8 and Fable 5?
Anthropic positions Sonnet 5 as delivering capability "close to that of Opus 4.8" at a lower price. [1][3] On the Artificial Analysis Intelligence Index, Sonnet 5's score of 53 sits roughly two to three points behind both Claude Opus 4.8 and GPT-5.5 at their highest effort settings, and further behind Claude Fable 5, which is built for the most demanding, long-horizon reasoning and agentic work. [3][5] Anthropic recommends Opus 4.8 as the default for complex agentic coding and enterprise work, Fable 5 for workloads needing the highest available capability, and Sonnet 5 for the best balance of speed, intelligence, and cost. [3]
The three models share a 1 million token context window and a 128,000-token maximum output, and none accepts manual extended-thinking budgets or non-default sampling parameters. [2][3] They differ mainly in price and peak capability: Fable 5 costs roughly three times as much per token as Sonnet 5, and Opus 4.8 sits between the two. [3]
How does Sonnet 5 relate to earlier Claude Sonnet models?
Sonnet 5 is a drop-in replacement for Claude Sonnet 4.6, continuing a line that runs through Claude Sonnet 4.5 and the original Claude 3.5 Sonnet. [2] Migrating requires only updating the model ID, but three behaviors changed: adaptive thinking is now on by default (it was off on Sonnet 4.6), manual extended thinking budgets return a 400 error, and setting temperature, top_p, or top_k to non-default values also returns a 400 error. [2] Sonnet 5 is additionally the first Sonnet-tier model with real-time cybersecurity safeguards, which may refuse prohibited or high-risk requests; such refusals return as a successful response with a refusal stop reason rather than an error. [2] Anthropic's safety testing found that Sonnet 5 shows an overall lower rate of undesirable behaviors than Sonnet 4.6 and is generally safer to use in agentic contexts. [1][4]
Where is Claude Sonnet 5 available?
At launch, Claude Sonnet 5 is available on the Claude API to all customers, in Claude Code and on claude.ai, and through Amazon Bedrock and Claude Platform on AWS, Google Cloud (Vertex AI), and Microsoft Foundry. [1][2] It supports zero data retention for organizations with the corresponding agreements. [2]
References
- Anthropic. "Introducing Claude Sonnet 5." June 30, 2026. https://www.anthropic.com/news/claude-sonnet-5 ↩
- Anthropic. "What's new in Claude Sonnet 5." Claude Platform Docs. https://platform.claude.com/docs/en/about-claude/models/whats-new-sonnet-5 ↩
- Anthropic. "Models overview." Claude Platform Docs. https://platform.claude.com/docs/en/about-claude/models/overview ↩
- Anthropic. "Claude Sonnet 5 System Card." June 30, 2026. https://www.anthropic.com/claude-sonnet-5-system-card ↩
- Artificial Analysis. "Claude Sonnet 5: Intelligence, Performance and Price Analysis." https://artificialanalysis.ai/models/claude-sonnet-5 ↩
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.