[AINews] Claude Opus 5: Fable-level performance at Opus price (half Fable)
Independent evals put Opus 5 ahead of Fable on agentic knowledge work, with lower cost per task.
Latent Space says Anthropic’s own messaging was restrained, describing Opus 5 as coming close to Fable even where official benchmarks had it technically ahead. The piece highlights Artificial Analysis ranking Opus 5 above Claude Fable 5 on AA-Briefcase by nearly 150 Elo, while cutting cost per task by 20%. Epoch’s numbers were more cautious: 159 ECI versus Fable 5’s 161, but a 161 SWE-ECI tie on software engineering. The note frames the launch reaction around benchmark friction, coding-agent praise, and early tool-use anecdotes. Latent Space's note
Latent Space says Anthropic’s own messaging was restrained, describing Opus 5 as coming close to Fable even where official benchmarks had it technically ahead. The piece highlights Artificial Analysis ranking Opus 5 above Claude Fable 5 on AA-Briefcase by nearly 150 Elo, while cutting cost per task by 20%. Epoch’s numbers were more cautious: 159 ECI versus Fable 5’s 161, but a 161 SWE-ECI tie on software engineering. The note frames the launch reaction around benchmark friction, coding-agent praise, and early tool-use anecdotes. Latent Space's note
score 9