Claude Sonnet 5

RawGraph

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]

SpecificationClaude Sonnet 5
DeveloperAnthropic
ReleasedJune 30, 2026
API model IDclaude-sonnet-5
Model tierSonnet (mid-tier, default)
PredecessorClaude Sonnet 4.6
Context window1,000,000 tokens (default and maximum)
Max output128,000 tokens (up to 300,000 via Batch API beta)
ModalitiesText and image input, text output
ReasoningAdaptive thinking, on by default
Knowledge cutoffJanuary 2026
Standard price$3 / $15 per million input / output tokens
Introductory price$2 / $10 per million tokens, through Aug 31, 2026

Sources: [1][2][3]

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]

BenchmarkClaude Sonnet 5Claude Sonnet 4.6GPT-5.5
SWE-bench Pro63.2%58.1%58.6%
Terminal-Bench 2.180.4%67.0%83.4%
OSWorld-Verified81.2%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

  1. Anthropic. "Introducing Claude Sonnet 5." June 30, 2026. https://www.anthropic.com/news/claude-sonnet-5
  2. Anthropic. "What's new in Claude Sonnet 5." Claude Platform Docs. https://platform.claude.com/docs/en/about-claude/models/whats-new-sonnet-5
  3. Anthropic. "Models overview." Claude Platform Docs. https://platform.claude.com/docs/en/about-claude/models/overview
  4. Anthropic. "Claude Sonnet 5 System Card." June 30, 2026. https://www.anthropic.com/claude-sonnet-5-system-card
  5. 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.

Suggest edit

What links here