What are the strengths of this model?

Extremely long 1000K context understanding Advanced conversational chat capability Latest design by Anthropic

What are the weaknesses of this model?

Non-open-source license Closed system with private internals Usage restrictions in closed environments

What are the best use cases?

Analyzing ultra-large documents Long-context conversational dialogues Building advanced AI chatbots

Back to Models

AnthropicProprietary

Claude Sonnet 4.5

Name: Claude Sonnet 4.5
Author: Anthropic

Claude Sonnet 4.5 is a foundation model developed by Anthropic. Designed for chat specialization, it supports an extremely extensive context window of 1000K.

Parameters

Undisclosed

Context Window

1000K

License

Proprietary

Release Date

2025-09-30

API Pricing

API pricing for this model is not yet available

Strengths

・Extremely long 1000K context understanding
・Advanced conversational chat capability
・Latest design by Anthropic

Weaknesses

・Non-open-source license
・Closed system with private internals
・Usage restrictions in closed environments

Use Cases

・Analyzing ultra-large documents
・Long-context conversational dialogues
・Building advanced AI chatbots

Deep Analysis

SWE-Bench Verified

77.2%

best at launch (Sept 2025)

GPQA Diamond

83.4%

strong reasoning

OSWorld

61.4%

SOTA computer use at launch

Input Price

$3/1M

same as Sonnet 4

Output Price

$15/1M

same as Sonnet 4

Context Window

200K tokens

1M beta at Tier 4+

Strengths

・Best coding model at launch (SWE-bench Verified 77.2%)
・Price parity with Sonnet 4 — pure capability upgrade
・Outstanding computer use (OSWorld 61.4%, 45% over Sonnet 4)
・Fast output speed (63 tok/s)

Weaknesses

・200K context window (1M in beta only)
・Superseded by Sonnet 4.6 and 4.8
・GPQA Diamond (83.4%) trails frontier models (94%+)

Competitor Comparison

Model	Arena	SWE	GPQA	Price
Claude Sonnet 4.6	~1490	79.6%	89.9%	$3/$15
GPT-5.4	1467	80.8%	94.4%	$2.50/$15
Claude Sonnet 4.5	~1460	77.2%	83.4%	$3/$15

Overview

Claude Sonnet 4.5 is Anthropic's mid-tier coding model, released on September 29, 2025. At launch, it achieved the highest SWE-bench Verified score (77.2%), establishing itself as the leading model for real-world software engineering tasks at its price point. The model maintains price parity with Sonnet 4 at $3/$15 per million tokens, making the upgrade a pure capability gain. It dominates computer use benchmarks with 61.4% OSWorld (45% improvement over Sonnet 4) and demonstrates exceptional agent capabilities with 98% on TAU-bench Telecom. Sonnet 4.5 has been superseded by Sonnet 4.6 (February 2026) and the upcoming Sonnet 4.8, which deliver further improvements. However, it remains a strong choice for cost-sensitive coding workloads.

Benchmarks & Performance

## Benchmark Scores | Benchmark | Sonnet 4.5 | Sonnet 4.6 | GPT-5.4 | |-----------|------------|------------|---------| | SWE-bench Verified | 77.2% | 79.6% | 80.8% | | GPQA Diamond | 83.4% | 89.9% | 94.4% | | OSWorld | 61.4% | ~65% | 75.0% | | TAU-bench Telecom | 98% | N/A | N/A | Sonnet 4.5 was a strong coding model at launch, with SWE-bench Verified leading all models. It has since been surpassed by Sonnet 4.6 and GPT-5.4.

Detailed Comparison

## Sonnet 4.5 vs Sonnet 4.6 Sonnet 4.6 delivers 2.4 points higher on SWE-bench Verified and 6.5 points higher on GPQA Diamond at the same price. Migration is recommended for improved reasoning and coding. ## Sonnet 4.5 vs GPT-5.4 GPT-5.4 offers a 5x larger context window (1M vs 200K) and higher benchmark scores across the board, at a lower price ($2.50/$15 vs $3/$15). GPT-5.4 is the better choice for most workloads.

Community Feedback

Sonnet 4.5 was praised for delivering best-in-class coding at the same price as Sonnet 4. Developers appreciated the pure upgrade without cost increases. The model's computer use capabilities (OSWorld 61.4%) were seen as a breakthrough for GUI automation tasks. However, the 200K context window was a limitation for long-document workflows. With the release of Sonnet 4.6 and 4.8, Sonnet 4.5 is now considered a legacy model but still performs well for coding tasks.

Use Cases

1. **Production Coding**: SWE-bench Verified 77.2% makes it suitable for code generation, review, and debugging in production environments. 2. **Computer Use Automation**: OSWorld 61.4% enables GUI testing, software automation, and visual workflow tasks. 3. **Agent Workflows**: TAU-bench Telecom 98% demonstrates exceptional performance on multi-step agent tasks. 4. **Budget-Conscious Teams**: At $3/$15, it offers strong coding capability at a lower price than frontier models.

Latest News

Claude Sonnet 4.5 was released September 29, 2025, and superseded by Claude Sonnet 4.6 on February 17, 2026. Sonnet 4.8 is expected in mid-2026. Sonnet 4.5 remains available via the API.

The model maintains price parity with Sonnet 4 at $3/$15 per million tokens, making the upgrade a pure capability gain. It dominates computer use benchmarks with 61.4% OSWorld (45% improvement over Sonnet 4) and demonstrates exceptional agent capabilities with 98% on TAU-bench Telecom.

Sonnet 4.5 has been superseded by Sonnet 4.6 (February 2026) and the upcoming Sonnet 4.8, which deliver further improvements. However, it remains a strong choice for cost-sensitive coding workloads.

Sources

Analysis generated: 2026-05-24