LLM Reference

Claude Sonnet 5

Released
2026-06-30
Last refreshed
2026-07-26
Status
Researched 41d ago
ProprietaryCommercial use: conditionalMultimodalCodingRAGAgentsLong contextVisionJSON / Tool useHighlight

Claude Sonnet 5 is worth evaluating for coding, rag, and agents when its provider route and context window match the workload.

Use it for

  • Teams evaluating coding, rag, and agents
  • Workloads that can use a 1m context window
  • Buyers comparing 4 tracked provider routes

Do not use it for

  • Workloads where another current model has stronger sourced task evidence
Specifications
Family
Claude 5
Released
2026-06-30
Context
1m
Max output
128,000
Architecture
Decoder Only
Knowledge cutoff
2026-01
Specialization
general
Openness
Proprietary
License
ProprietaryCommercial use: conditional
Weights
Not released
Code
Unknown
Training
Fine-tuned
Created by

Developing safe and ethical AI systems.

San Francisco, California, United States
Founded 2021
Website
Pricing
Output / 1M
$10.00
Input / 1M
$2.00

Cheapest of 5 routes · Anthropic · cache read $0.200

About

Claude Sonnet 5 is Anthropic's next-generation Sonnet model for agentic coding, tool use, computer use, and professional work. It is a proprietary decoder-only model with a 1M-token context window, 128K max output, multimodal vision, adaptive thinking, function calling, structured outputs, prompt caching, and Batch API support. It is available through the Claude API, AWS Bedrock, Google Cloud Vertex AI, Microsoft Foundry preview, and OpenRouter. Anthropic lists durable standard pricing at $3/1M input and $15/1M output tokens, with introductory $2/$10 pricing through 2026-08-31.

Claude Sonnet 5 is a proprietary model in the Claude 5 family. The structured metadata tracks a 1m-token context window, multimodal input, reasoning, function calling, tool use, structured outputs, and code execution. This page tracks provider routes through Anthropic, AWS Bedrock, GCP Vertex AI, and 2 more, with the cheapest tracked route listed at $2 input and $10 output per 1M tokens. Headline tracked benchmarks include SWE-bench Verified 85.2, SWE-bench Pro 63.2, and SWE-bench Multilingual 78.3.

Top use-case fit: coding, agents, and build tasks

Coding

Q/$ D

2 relevant benchmarks in the decision map.

RAG

Included by capability and metadata signals in the decision map.

Agents

Q/$ D

1 relevant benchmark in the decision map.

Provider price ladder

Compare all 5

Compare API pricing across 4 providers for input and output tokens, batch, and cached reads when available.

ProviderInput / 1MOutput / 1MBatch in / outCacheRoute
Anthropic$2.00$10.00$1.00 / $5.00read $0.200 / 5m $2.50 / 1h $4.00
Serverless
GCP Vertex AI$2.00$10.00$1.00 / $5.00read $0.200 / 5m $2.50 / 1h $4.00
Serverless
OpenRouter$2.00$10.00-read $0.200 / 5m $2.50 / 1h $4.00
Serverless
AWS Bedrock----
ServerlessPartial

Available via routers & gateways(16)

Capabilities

VisionMultimodalReasoningFunction CallingTool UseStructured OutputsCode ExecutionPrompt CachingBatch API

Benchmark peer barsfor Coding

Benchmark scores(16)

Scores are benchmark-specific and are direction-aware: the same numeric gap can mean very different outcomes across suites. Use the leaderboard context and this model's provider route to decide whether the winning margin is meaningful for your workload.
BenchmarkScoreVersionEvaluationSource
SWE-bench Verified85.2SWE-bench Verified; Anthropic standard configuration, adaptive thinking at max effort, default sampling, average over 5 trialsObserved 2026-06-30Source
SWE-bench Pro63.2SWE-bench Pro; Anthropic standard configuration, adaptive thinking at max effort, default sampling, average over 5 trialsObserved 2026-06-30Source
SWE-bench Multilingual78.3SWE-bench Multilingual; Anthropic standard configuration, adaptive thinking at max effort, default sampling, average over 5 trialsObserved 2026-06-30Source
Terminal-Bench 2.180.4Terminal-Bench 2.1; mini-SWE-agent harness on GKE, 1x timeout rate, 3x memory ceiling, xhigh effort, 445 trialsObserved 2026-06-30Source
Humanity's Last Exam — With Tools57.4Humanity's Last Exam with web search, web fetch, programmatic tool calling, and code execution; Claude Opus 4.6 grader; 1M total-token capObserved 2026-06-30Source
BrowseComp — Multi-Agent86.6BrowseComp multi-agent; web search, web fetch, programmatic tool calling, code execution; 10M-token limit with context compactionObserved 2026-06-30Source
BrowseComp84.7BrowseComp single-agent; adaptive thinking at maximum effort, 10M-token limit with context compaction triggered at 200kObserved 2026-06-30Source
OSWorld-Verified81.2OSWorld-Verified; 361 tasks, 1080p Ubuntu VM, 100 action steps, adaptive thinking at max effort, pass@1 averaged over 5 runsObserved 2026-06-30Source
GDP.pdf81.6GDP.pdf with Python tools and image cropping tool; internal harness, Opus 4.7 judge, mean criteria pass rate averaged over 5 runsObserved 2026-06-30Source
Humanity's Last Exam43.2Humanity's Last Exam; no tools; reasoning-only; thinking auto; 1M total-token cap; Claude Opus 4.6 graderObserved 2026-06-30Source
CursorBench61.2CursorBench 3.1Observed 2026-06-30
Configuration: Sonnet 5 Max
Harness: CursorBench 3.1
Evaluator: Cursor
Confidence: confirmed
Notes: Highest CursorBench 3.1 score across Cursor's published effort configurations for this base model.
Source
CursorBench61.5CursorBench 3.2Observed 2026-07-18
Configuration: Sonnet 5 Max
Harness: CursorBench 3.2 productized Cursor-agent workflow
Evaluator: Cursor
Confidence: confirmed
Cost/task: $6.45
Tokens/task: 92,882
Steps/task: 86
Notes: Cursor vendor-reported result; not independently reproducible. Results are subject to variance, and small score differences may not be statistically meaningful.
Source
CursorBench58.7CursorBench 3.2Observed 2026-07-18
Configuration: Sonnet 5 Extra High
Harness: CursorBench 3.2 productized Cursor-agent workflow
Evaluator: Cursor
Confidence: confirmed
Cost/task: $4.16
Tokens/task: 52,871
Steps/task: 67
Notes: Cursor vendor-reported result; not independently reproducible. Results are subject to variance, and small score differences may not be statistically meaningful.
Source
CursorBench56.9CursorBench 3.2Observed 2026-07-18
Configuration: Sonnet 5 High
Harness: CursorBench 3.2 productized Cursor-agent workflow
Evaluator: Cursor
Confidence: confirmed
Cost/task: $3.19
Tokens/task: 39,483
Steps/task: 57
Notes: Cursor vendor-reported result; not independently reproducible. Results are subject to variance, and small score differences may not be statistically meaningful.
Source
CursorBench52.4CursorBench 3.2Observed 2026-07-18
Configuration: Sonnet 5 Medium
Harness: CursorBench 3.2 productized Cursor-agent workflow
Evaluator: Cursor
Confidence: confirmed
Cost/task: $2.16
Tokens/task: 26,200
Steps/task: 46
Notes: Cursor vendor-reported result; not independently reproducible. Results are subject to variance, and small score differences may not be statistically meaningful.
Source
CursorBench47.7CursorBench 3.2Observed 2026-07-18
Configuration: Sonnet 5 Low
Harness: CursorBench 3.2 productized Cursor-agent workflow
Evaluator: Cursor
Confidence: confirmed
Cost/task: $1.30
Tokens/task: 16,269
Steps/task: 33
Notes: Cursor vendor-reported result; not independently reproducible. Results are subject to variance, and small score differences may not be statistically meaningful.
Source

Migration checks

No linked migration route is available for this model yet.

Frequently asked questions

What is the context window of Claude Sonnet 5?

Claude Sonnet 5 has a context window of 1m tokens.

What is the max output of Claude Sonnet 5?

Claude Sonnet 5 can generate up to 128,000 output tokens.

How much does Claude Sonnet 5 cost?

Claude Sonnet 5 pricing ranges from $2.00/1M to $2/1M input tokens depending on the provider.

When was Claude Sonnet 5 released?

Claude Sonnet 5 was released on 2026-06-30.

Which providers offer Claude Sonnet 5?

Claude Sonnet 5 is available from 5 providers: Anthropic, AWS Bedrock, GCP Vertex AI, OpenRouter, Microsoft Foundry.

What benchmarks has Claude Sonnet 5 been tested on?

Claude Sonnet 5 has been evaluated on 16 benchmarks, including SWE-bench Verified, SWE-bench Pro, SWE-bench Multilingual, Terminal-Bench 2.1, Humanity's Last Exam — With Tools.