Claude Opus 4.8

Released 2026-05 · hybrid · 1.0M tokens · 5 benchmarks

Editorial notes

Lanzado 28 mayo 2026. Flagship GA. Upgrade directo a mismo precio $5/$25 per MTok (fast mode opcional $10/$50, 3x mas barato que prior fast). 1M context, 128K max output, text+image, knowledge cutoff enero 2026 (model id anthropic.claude-opus-4-8). Nuevo: parallel-subagent dynamic workflows en Claude Code, mid-task system messages en Messages API, user-controllable effort level. Scores del System Card oficial (extraidos via Vellum/llm-stats — el PDF excedio fetch limit; identity/specs primary-confirmed via anthropic.com + AWS Bedrock). Otros scores System Card: Terminal-Bench 2.1 74.6, HLE con tools 57.9, Online-Mind2Web 84, Finance Agent v2 53.9. Caveat: GPQA Diamond 93.6 es el unico bench donde 4.8 va ligeramente debajo de Opus 4.7 (94.2).

Spec sheet

Empresa: Anthropic
Pais: US
Tipo: hybrid
Release: 2026-05
Context: 1.0M tokens
Slug: claude-opus-4-8

Benchmarks (5)

Reasoning 2

Coding 2

Agentic 1

OSWorld

Computer use benchmark - tareas reales de escritorio.

83.4

Cite this model

BibTeX · APA

BibTeX

@misc{frontier-claude-opus-4-8,
  title  = {Claude Opus 4.8},
  author = {{Anthropic}},
  year   = {2026},
  note   = {Frontier Benchmarks AI atlas. Accessed 2026-06-10},
  url    = {https://frontierbenchmarks.com/models/claude-opus-4-8}
}

APA

Anthropic (2026). Claude Opus 4.8 [Large language model]. Frontier Benchmarks AI. Retrieved 2026-06-10, from https://frontierbenchmarks.com/models/claude-opus-4-8

Citation refleja la pagina del atlas, no el paper original del modelo. Para el paper, ve a la seccion "Recursos" arriba.

⚔️ Battle vs otro modelo ← Todos los modelos Mas de Anthropic Methodology