Claude Sonnet 4.5 vs Claude Sonnet 5

Pricing math

Provider list price: Sonnet 4.5 runs $3/M input and $15/M output; Sonnet 5 runs $2/M input and $10/M output — a 1.5x gap on input pricing. On OpenKey, add the flat 3% fee: Sonnet 4.5 becomes $3.09/M input and $15.45/M output (3.00 x 1.03, 15.00 x 1.03); Sonnet 5 becomes $2.06/M input and $10.30/M output (2.00 x 1.03, 10.00 x 1.03). Run the numbers on a 10M-input / 2M-output workload and Sonnet 4.5 costs $60.00 while Sonnet 5 costs $40.00 — a $20 difference on that single job. Cache pricing follows the same ratio: Sonnet 4.5 charges $0.30/M cache-read and $3.75/M cache-write versus Sonnet 5's $0.20/M and $2.50/M. Both models bill on the same OpenKey key with the same 3% fee, so the pricing gap here is pure provider pricing, not markup.

Coding and agent performance

Sonnet 4.5 has a full Design Arena benchmark history: rank 34 in codecategories (elo 1232, 53% win rate), rank 34 in 3d (elo 1236), rank 10 in asciiart (elo 1249, 56.8% win rate), and agent-arena results including rank 18 in fullstack (elo 1120, 43.5% win rate), rank 21 in mobileapps (elo 1184, 48.9% win rate), and rank 21 in webapps (elo 1129, 43.1% win rate). Sonnet 5 doesn't have Design Arena entries yet, but Artificial Analysis puts it at a 71.5 coding index, 53.4 intelligence index, and 46.7 agentic index. These come from different measurement systems, so you can't directly rank one against the other — treat Sonnet 4.5's numbers as proven track record and Sonnet 5's as a newer, differently-scored data point.

Output length and reasoning control

Sonnet 5 doubles the max completion length: 128,000 tokens versus Sonnet 4.5's 64,000. That matters for long-form generation, large diffs, or multi-file code output in a single response. Sonnet 5 also exposes selectable reasoning effort — low, medium, high, xhigh, or max, defaulting to medium — while Sonnet 4.5 only supports reasoning as an on/off toggle without effort tuning. If your workload needs to dial reasoning depth per request instead of eating a fixed cost every call, Sonnet 5's parameter set (verbosity, max_completion_tokens, effort levels) gives you more control.

When to pick each

Pick Sonnet 4.5 when you need a model with an established Design Arena benchmark record across 3D, SVG, game dev, UI components, and web apps — useful if you're citing hard evidence in a technical decision. Pick Sonnet 5 for cost-sensitive production workloads, long-output tasks, or anything where tunable reasoning effort saves you money on simpler calls. Both share the same 1M context window and file/image input support, so context size and modality aren't differentiators here — cost, output cap, and reasoning control are.

Questions

How much cheaper is Claude Sonnet 5 than Sonnet 4.5?

On provider list pricing, Sonnet 5 is $2/M input and $10/M output versus Sonnet 4.5's $3/M and $15/M — a 1.5x ratio on input cost. On a 10M-input/2M-output workload, that's $40.00 for Sonnet 5 versus $60.00 for Sonnet 4.5, a $20 difference.

Do both models have the same context window?

Yes. Both Claude Sonnet 4.5 and Claude Sonnet 5 support a 1,000,000-token context window — a 1.0 context ratio between them. The difference is in max output: 64,000 tokens for Sonnet 4.5 versus 128,000 for Sonnet 5.

Which model has published coding benchmarks?

Claude Sonnet 4.5 has Design Arena scores across 11 categories, including a codecategories elo of 1232 (rank 34, 53% win rate). Sonnet 5 instead reports an Artificial Analysis coding index of 71.5 — a different scoring system, so the two aren't directly comparable.

What's the OpenKey price difference including fees?

With the flat 3% fee, Sonnet 4.5 costs $3.09/M input and $15.45/M output on OpenKey (3.00 x 1.03 and 15.00 x 1.03). Sonnet 5 costs $2.06/M input and $10.30/M output (2.00 x 1.03 and 10.00 x 1.03). Both run through the same OpenKey key.