The Latent
EN▾
EnglishEspañol中文PortuguêsFrançaisالعربية日本語한국어
Sign Up
NEWSDATARESEARCHPODCASTNEWSLETTERPRESSPARTNER WITH US →
Data/Capability

Capability

Agentic and Tool-Use Benchmarks

Agentic and Tool-Use Benchmarks. Demo series compiled by The Latent research desk.

Agentic and Tool-Use Benchmarks

  • SWE-bench Verified
  • OSWorld
  • GAIA
Agentic and Tool-Use BenchmarksDemo preview through July 2026 (percent).100%75%50%25%0%Aug '25Oct '25Dec '25Feb '26Apr '26Jun '26SWE-bench Verified: 77.6% on Jul '26. OSWorld: 51.3% on Jul '26. GAIA: 68.8% on Jul '26
SOURCE: The LatentUPDATED: Jul 1, 2026
ZOOMALLYTD12M3M1M
CSV
View data
DateSWE-bench VerifiedOSWorldGAIA
2025-08-0176.7750.7267.99
2025-09-0181.5753.9572.39
2025-10-0176.4150.5767.88
2025-11-0176.3250.4767.7
2025-12-0175.249.7666.79
2026-01-0172.4147.8764.2
2026-02-0179.9652.8370.82
2026-03-0182.0454.2672.8
2026-04-0182.2354.3672.91
2026-05-0185.2956.4275.71
2026-06-0178.5952.0269.85
2026-07-0177.651.368.8

API access · Pro coming soon

Demo preview through July 2026 (percent).

Download CSVPro API coming soon

Methodology

Synthetic demo series generated to preview the data product layout.

Related charts

  • Historical Token Throughput per GPU
  • Historical Inference Speeds
  • Historical Token Efficiency Across Frontier Models

The Latent

AI industry news. A sister publication to The Block.

Editorial

  • Standards
  • Corrections
  • Commercial policy
  • Contact

Company

  • About
  • Privacy Policy
  • Terms of Service
  • Security
  • The Block