Anthropic · language model
Claude Sonnet 4.5: pricing, context and benchmarks
Claude Sonnet 4.5 is Anthropic's model released on September 29, 2025. It costs $3 per million input tokens and $15 per million output tokens, with a 1M-token context window. Epoch AI ranks it #72 of 125 rated models.
| Performance (ECI) | 17/40 | |
|---|---|---|
| Price | 3/30 | |
| Context | 12/15 | |
| Features | 11/15 |
What you need to know about Claude Sonnet 4.5
Anthropic, based in United States, put Claude Sonnet 4.5 online on September 29, 2025. It accepts text, images, and files as input and produces text. Its context window reaches 1,000,000 tokens, larger than 77 % of the models in the registry, and it can write up to 64,000 tokens in one go.
It has been available for about 13 months, long enough for its prices and scores to have settled.
Claude Sonnet 4.5 pricing
On OpenRouter, Claude Sonnet 4.5 costs $3 per million input tokens and $15 per million output tokens. Counting three tokens read for every token written, its average price is $6 per million, more expensive than 91 % of paid models (7.9 times the median of $0.76).
A concrete example: a task that sends 2,000 tokens (about 1,500 words) and receives a 500-token answer costs $0.0135. At 1,000 tasks a day, budget $405 a month. The price calculator redoes this with your own volumes.
Tokens already sent and cached are billed at $0.3 per million. With batch processing (delayed answers), the price drops to $1.5 input and $7.5 output.
Performance and benchmarks
The Epoch Capabilities Index (ECI), which aggregates published results across many benchmarks, gives Claude Sonnet 4.5 a score of 146.7 (range 145 to 148.3). That places it #72 of 125 rated models.
Scores published by Epoch AI on the benchmarks the site tracks:
| Benchmark | What it measures | Score | Date |
|---|---|---|---|
| GPQA Diamond | PhD-level science questions | 82.3 % | Sep 29, 2025 |
| SWE-bench Verified | fixing real GitHub issues | 71.3 % | Sep 29, 2025 |
| Terminal-Bench | real tasks in a terminal | 46.5 % | Sep 29, 2025 |
| ARC-AGI-1 | visual abstract-reasoning puzzles | 63.7 % | Sep 29, 2025 |
| ARC-AGI-2 | harder version of the ARC puzzles | 13.6 % | Sep 29, 2025 |
| FrontierMath | research-level maths problems (tiers 1-3) | 23.9 % | Sep 29, 2025 |
| OTIS Mock AIME | maths competition problems | 77.8 % | Sep 29, 2025 |
| SimpleQA Verified | accuracy on simple facts | 30.7 % | Sep 29, 2025 |
| OSWorld | operating a desktop computer | 62.9 % | Sep 29, 2025 |
For at least equal performance, GPT-6.1 Sol, Gemini 4 Argon, and Claude Sonnet 5.5 cost at least 30% less.
Features, privacy and local use
- It supports tool calling, so it can drive an AI agent, call APIs or MCP servers.
- It can answer in an enforced JSON format (structured outputs).
- It is a reasoning model: it "thinks" before answering, which helps on hard tasks but uses more output tokens.
- Its knowledge stops at January 2025.
Its weights are not published: it can only be used through the Anthropic API or a broker such as OpenRouter, which means sending your data off your machines.
Other Anthropic models
| Model | Released | Avg. price | ECI |
|---|---|---|---|
| Claude Haiku 5.5 | Oct 7, 2026 | $0.2 | 159.3 |
| Claude Sonnet 5.5 | Sep 28, 2026 | $4 | 166.2 |
| Claude Opus 5.5 | Sep 22, 2026 | $8 | 168.8 |
| Claude Fable 5.1 | Sep 1, 2026 | $20 | 166.5 |
| Claude Opus 5 | Jul 24, 2026 | $10 | 163.9 |
| Claude Sonnet 5 | Jun 30, 2026 | $4 | 156.9 |
| Claude Fable 5 | Jun 9, 2026 | $20 | 163.4 |
| Claude Opus 4.8 | May 27, 2026 | $10 | 159 |
Verdict
Computed score: 43/100, from public data (performance, price, context, features).
Strengths
- Very large context (1M tokens): long documents or a whole codebase fit in one request.
- Agent-ready: tool calling and structured JSON output.
- Discounted batch pricing.
Limitations
- Expensive: costlier than 91 % of paid models.
Avoid if…
- you process large volumes on a tight budget;
- your data must not leave your servers;
- your tasks are mostly simple: a non-reasoning model will cost less.
Alternatives
Comparisons and tools
Learn more
Frequently asked questions
How much does Claude Sonnet 4.5 cost?
$3 per million input tokens and $15 per million output tokens on OpenRouter (checked October 25, 2026). A task with 2,000 tokens read and 500 written costs $0.0135.
How big is Claude Sonnet 4.5's context window?
1,000,000 tokens, roughly 750,000 words, with a maximum of 64,000 output tokens.
Can Claude Sonnet 4.5 run locally?
No, Anthropic does not publish its weights: it is API-only.
Is Claude Sonnet 4.5 suitable for an AI agent?
Technically yes: it supports tool calling and structured outputs. Then check its cost for your volume.
How good is Claude Sonnet 4.5?
Epoch AI ranks it #72 of 125 rated models (ECI 146.7). On GPQA Diamond, it scores 82.3%.
What are the alternatives to Claude Sonnet 4.5?
The closest models on price and performance are GPT-5.1, GPT-5, and o3.
Related articles
- Mistral AI: models, pricing and agentsOctober 25, 2026
- LLM benchmarks: how to read model rankingsOctober 24, 2026
- Running a local LLM: hardware, software, and limitationsOctober 24, 2026
- Open-source or closed models for your AI agents?October 5, 2026
Neutrality. No affiliate links. No vendor pays to be reviewed, rated or ranked. Reviews are directed by François Chollet, neutral and factual, and every data point is dated and sourced. How we score