Artificial Analysis places Claude Sonnet 4.6 (Non-reasoning, High Effort) among comparable non-reasoning models at Intelligence Index 37, approximately 46 tok/s, input $3 / output $15 per million tokens, with a stated 1M context; the page also notes this model is deprecated, and the intelligence score no longer represents the latest Sonnet.
Display configuration: Claude Sonnet 4.6, Non-reasoning, Effort high.
Index version: Artificial Analysis Intelligence Index v4.1.1, including GDPval-AA v2, τ³-Banking, Terminal-Bench v2.1, SciCode, Humanity's Last Exam, GPQA Diamond, CritPt, AA-Omniscience, AA-LCR.
Comparison pool: The page states non-reasoning models are compared only against other non-reasoning models; pricing tiers are compared against proprietary models in the same category with >$1/1M blended.
Maintenance status: This model is deprecated. Only continues default 10k input token load performance benchmarks; results for other loads are historical values. Suggests considering Claude Sonnet 5 (Non-reasoning).
The page shows the non-reasoning + high effort variant, not the reasoning variant. Input modalities text+image, output text, context 1M. Specific per-benchmark prompts and harness are in AA's Intelligence Index methodology, not fully expanded on this model page.
Model summary at collection time:
| Metric | Page value |
|---|---|
| Intelligence Index | 37 (category #5 / 63; category median 23) |
| Speed | 46.1 output tok/s (category #36 / 63; page says notably slow) |
| Input price | $3.00 / million tokens |
| Output price | $15.00 / million tokens |
| Cache discount | 90% |
| Cost per Intelligence Index task | N/A |
| Verbosity | N/A |
| Context | 1M tokens |
Same-page Intelligence comparison bar (excerpt, higher is better): Claude Opus 5 (max) 63, Claude Fable 5 62, GPT-5.6 Sol (max) 61, …, Claude Sonnet 4.6 (Non-reasoning) 37. In the speed comparison, Sonnet 4.6 is 46 tok/s, slower than most listed comparison models.
Page copy: Among the leading non-reasoning models on intelligence, but relatively expensive for same-price-tier non-reasoning models; supports text/image input.
By AA's methodology, Sonnet 4.6 non-reasoning high effort remains clearly stronger than the category median, but is no longer at the intelligence frontier as of 2026-08, and speed is also on the slow side. Suitable as a production reference point for "known unit price, 1M window, non-reasoning," but not suitable as a proxy for the latest Sonnet anymore. When reasoning variant scores are needed, open AA's reasoning page; do not substitute this page's 37-point score.
Page explicitly deprecated; some workloads no longer updated.
This page is Non-reasoning High Effort, not interchangeable with official default effort or results with thinking enabled.
Cost per task and verbosity are N/A; cannot infer per-task dollars from this page.
Intelligence Index is a composite of 9 items; cannot be decomposed into SWE or OSWorld.
Fix AA's v4.1.1 methodology, explicitly choosing whether to run non-reasoning or reasoning.
Set claude-sonnet-4-6 with effort=high, disable reasoning.
Record Index subscores, tok/s, latency, and actual $3/$15 billing separately.
When comparing against Sonnet 5 non-reasoning, use AA's currently maintained pages, not this deprecated page's historical loads.
Claude Sonnet 4.6
