Skip to content

Claude Sonnet 5.5

AnthropicProprietaryReasoningVerifiedReleased Sep 28, 2026Input: text, image, file
Compare with Gemini 4 ArgonSource scores ↗OpenRouter ↗
Overall score
82.2
Rank 6 of 222 · 76.5–88.0
Context window
1M
1,000,000 tokens
Price per 1M tokens
$2.00 in · $10.00 out
✓ Verified on Anthropic's pricing page
Blended price
$4.00
3:1 input:output tokens
Output speed
—
Not measured yet

Category scores

Scores per capability area with the model's rank among ranked models in that category.

  1. 1Math95.2
  2. 2Coding#187.5
  3. 3Multimodal86.7
  4. 4Knowledge#583.4
  5. 5Reasoning#681.8
  6. 6Agentic & tool use#1178.8
Higher is better. #N is the model's rank in that category, where it is eligible to be ranked.

Benchmark results (57)

As reported in system cards, papers and public evaluations. Position compares this model with every other model reporting the same benchmark.

Knowledge

HLE w/o tools56.9#4 of 43
HealthBench Professional69.2#1 of 16
HealthBench (raw)69.4#1 of 7
HealthBench (length-adjusted)65.4#1 of 7
HealthBench Professional (raw)77.1#1 of 7
BioMysteryBench (human-difficult)44.7#4 of 6
BioMysteryBench (human-solvable)89.2#3 of 5
SpatialBench Verified72.5#1 of 3
SingleCellBench59.1#3 of 3
Protein Design51.0#2 of 3
Protocols (troubleshooting)67.3#2 of 3
Protocols (understanding)66.6#3 of 3
Morphology-to-molecule matching25.0#2 of 2
Medicinal chemistry65.3#1 of 2
Protein Design library ranking54.8#2 of 2
De novo protein-binder design82.3#2 of 2

Agentic & tool use

MazeBench8#9 of 42
HLE w/ tools64.5#3 of 27
Toolathlon-Verified77.8#3 of 22
Terminal-Bench 4.070.60#1 of 18
Hermes Index53.1#4 of 16
Hermes Bench73.5#6 of 16
DRACO87.0#2 of 6
Terminal-Bench-Science 0.159.9#2 of 5
Toolathlon Verified Pass@385.2#2 of 4
Toolathlon Verified Pass³68.5#4 of 4
Toolathlon Verified avg. turns(lower is better)31.6#4 of 4
LAB all-pass (Harvey held-out)10.0#2 of 3
LAB criterion-pass (Harvey held-out)93.1#2 of 3
AutomationBench (Zapier 1.0.6)44.70#1 of 1

Multimodal

Design Arena Website1324#5 of 91
ParseBench70.18#13 of 37
ExtractBench (LlamaIndex)(lower is better)94.68#24 of 24
OfficeQA Pro65.6#5 of 13
SWE-bench Multimodal54.3#5 of 7
ExtractBench Pro32#3 of 5
Chartography (no tools)61.6#3 of 5
Chartography (tools)90.2#1 of 5
BenchCAD Vision2Code (tools)0.963#1 of 4
BenchCAD Vision2Code (no tools)0.747#1 of 3
OfficeQA76.9#3 of 3
Biomedical image analysis72.2#1 of 2

Coding

SWE-bench Pro81.3#2 of 69
DeepSWE71.0#16 of 50
Bug Hunt Bench51.3#1 of 23
FrontierSWE v261.9#3 of 20
FrontierCode 1.1 Main46.2#8 of 16
ProgramBench79.7#5 of 15
FrontierCode 1.1 Extended59.1#6 of 8

Math

FrontierMath v2 (Tiers 1-3)88.772#7 of 64
FrontierMath v2 (Tier 4)80.488#8 of 51
ArXivMath Aug. 2026 (no tools)86.8#2 of 2
ArXivMath Aug. 2026 (tools)95.2#2 of 2

Multilingual

SWE Multilingual90.3#3 of 46
GMMLU92.1#3 of 3
MILU91.6#3 of 3

Pricing by source

USD per 1M tokens.

SourceInputOutputCached in
✓ Official (Anthropic)
checked Oct 17, 2026 · Claude Sonnet 5.5
$2.00$10.00$0.10
BenchLM list price$2.00$10.00$0.10
OpenRouter anthropic/claude-sonnet-5.5$2.00$10.00$0.10

Anthropic’s current pricing docs list $2 per million input tokens, $10 per million output tokens, $0.10 cache reads, $2.50 five-minute cache writes, and $4 one-hour cache writes. The October 7, 2026 Haiku 5.5 announcement cut Sonnet 5.5 cache reads by 50% from $0.20. Batch API rates are $1 input and $5 output per million tokens; the cached-input batch estimate is half the current cache-read rate. Anthropic states that the 1M context window is included at standard pricing.

Arena ratings

Human-preference ratings from LMArena. Best-rated variant shown.

Text(xhigh)1470#30 · 6.8K votes
WebDev (coding)(xHigh)1768#3 · 3.4K votes
Vision(xHigh)1287#30 · 2.6K votes

Speed

Measured by us through OpenRouter (median of trials).

Not measured yet.

Related models

Same family

Similar score