Anthropic launches Opus 5 as efficient alternative to pricier Fable model

Anthropic has unveiled Opus 5, the latest iteration in its Claude model family, aiming to deliver strong performance on software development work without the premium price tag of its more advanced alternatives.

Performance gains focused on incremental improvement#

Rather than represent a dramatic leap forward, Opus 5 demonstrates steady progress across key benchmarks. Industry tests including Frontier-Bench and DeepSWE position the model at comparable or slightly better levels than Anthropic's Fable model for coding, while consistently outpacing both Opus 4.8 and OpenAI's GPT-5.6-Sol across most task categories. The gains, however, are evolutionary rather than transformative—advancing incrementally on each metric without fundamentally reshaping what the model can do.

Cost-performance trade-off#

The core appeal lies in the economics: Opus 5 reaches near-Fable-tier performance at approximately half the operational cost. This positioning targets organizations seeking sophisticated coding capabilities without committing to premium pricing.

Deliberate limitations in cybersecurity#

The company made intentional design choices to dial back Opus 5's security-focused training. As a result, it lags substantially behind both Fable and Mythos on identifying and exploiting cybersecurity vulnerabilities, though Anthropic maintains the model still performs reasonably at vulnerability discovery. Unlike Fable, Opus 5 also avoids certain safety measures, such as Fable's 30-day data retention policy for incident review.

Learn more: Anthropic's announcement