readable AI benchmarks (simplified)

readable AI benchmarks

Claude Sonnet 5 (Low) has a Quality Score of 41.6, ranking 175th among 571 scored models. Per 1M tokens, pricing is $2.00 input, $10.00 output, $0.200 cached input, and $2.50 cache write. It has a 1M token context window. Its Reliability Score is 45.9, ranking 170th among 573 scored models, above average (total average: 36.5). Its Value Score is 27.6, around average compared with the total Value average of 31.4.

Quality Score41.6coverage 100.0%: AA Intelligence and AA-Omniscience outcomes available

Value Score27.6

Factual reliability45.9

Cache Discount90%

Model specification

Reasoning: reasoning
Input modalities: none
Output modalities: none
Context window: 1M tokens
Weights: closed weights

Token prices USD per 1M tokens

Input: $2.00
Output: $10.00
Cached Input: $0.200
Cache write: $2.50
Cost per task: $0.509

Capability

Intelligence: 24.3
Coding: not available
Agentic: not available
Omniscience: -8.2
Correct: 37.4%
Blended price: $1.54

Answer outcomes

Fully graded outcomes (Correct + Incorrect): 82.9%

Correct 37.4%Incorrect 45.6%Partial / not attempted 17.1%

Artificial Analysis benchmarks

GDPval-AA v2: 29%
SciCode: 50%
Humanity’s Last Exam: 22%
CritPt: 5%
AA-Omniscience: 46%
AA-LCR: 67%

DeepSWE operational cross-check

Pass@1: 30.5%
Avg steps/scored attempt: 76.89
Mean cost/scored attempt (USD): $2.19
Mean output tokens/scored attempt: 35,595.082
Mean duration/scored attempt (s): 735 s

Mean cost/scored attempt is the official feed's mean_cost_usd: mean benchmark cost in USD per scored attempt under the source's cost basis. It is not Artificial Analysis cost per task and not token price (USD per 1M tokens). Mean output tokens/scored attempt is mean_output_tokens, also averaged over every scored attempt including scored failures. Mean duration/scored attempt is mean_duration_seconds; it is informational only, and wall-clock varies with provider and host load. These three fields do not affect Quality Score, Value Score, or Reliability. Epoch General ECI's benchmark basket includes DeepSWE v1.1, so Epoch General ECI and this DeepSWE cross-check are not independent evidence.

official DeepSWE v1.1 source

Similar models

Graph Open on Artificial Analysis