Model
Model explorer

Claude Sonnet 5

CLOSED
Anthropic · Claude 5 family · released Jun 30, 2026

Model id claude-sonnet-5. Drop-in upgrade for Sonnet 4.6: adaptive thinking on by default, manual extended-thinking removed (400 error), new tokenizer (~30% more tokens for same text vs Sonnet 4.6), non-default sampling params rejected. Introductory pricing $2/$10 per MTok through 2026-08-31, standard pricing $3/$15 thereafter (used here). On Terminal-Bench 2.1 and GDPval-AA v2 it narrowly beats Opus 4.8 despite Sonnet-tier pricing. NO SEPARATE '(High)' PRODUCT: Anthropic's docs state explicitly that setting effort to 'high' 'produces exactly the same behavior as omitting the effort parameter entirely,' and high is the documented default — there is no distinct marketed 'Sonnet 5 (High)' SKU to add as a sibling entry.

ReasoningCodingVisionFunction callingTool useAgentic
2490.3
Elo · rank #19
Parameters
Undisclosed
Active params
Context
1M tokens
Architecture
Dense transformer, architecture undisclosed; adaptive thinking on by default, controlled by a runtime effort parameter (low/medium/high/xhigh/max)
License
Proprietary
Languages
API price (in/out)
$3 / $15
Modalities
text · vision
Benchmark results
Bar shows position within the tracked field; marker = field best
AIMEMath86.7%#59
best: GPT-5.2 · 100.0%
BrowseCompAgents84.7%#6
best: Kimi K3 · 91.2%
CharXivVision77.0%#18
best: Muse Spark 1.1 (xhigh) · 88.4%
GDPval-AAAgents1618#8
best: Claude Fable 5 · 1932
GPQA DiamondReasoning90.5%#20
best: GPT-5.6 · 94.6%
Humanity's Last ExamReasoning57.4%#1
best: this model · 57.4%
LiveCodeBenchCoding82.4%#37
best: DeepSeek-V4-Pro (Think Max) · 93.5%
OSWorld-VerifiedAgents81.2%#3
best: Claude Fable 5 · 85.0%
SWE-bench ProCoding63.2%#6
best: Claude Fable 5 · 80.0%
SWE-bench VerifiedCoding85.2%#5
best: Claude Fable 5 · 95.0%
Terminal-Bench 2.0Coding80.4%#8
best: GPT-5.6 · 88.8%
Run it locally
Closed weights — available via API only. No local deployment.
Input / M tok
$3
Output / M tok
$15
Claude 5 family
Elo progression across releases
API price $3/$15 · each benchmark row carries its own source badge (see methodology)