Skip to content
LLMAnthropic

Claude Opus 4 redefines what language models can do

Manu Ramalho
Manu Ramalho

Anthropic launched Claude Opus 4, the most capable model ever created by the company, with significant advances in complex reasoning and code generation.

Loading audio…
Cover image for: Claude Opus 4 redefines what language models can do

Anthropic announced today the release of Claude Opus 4, marking a significant leap in the ability of language models. The new model brings substantial improvements in three key areas: complex reasoning, code generation and long instruction tracking.

What changes with Opus 4

The main differential of Opus 4 is its ability to maintain coherent reasoning chains by thousands of tokens. In internal benchmarks, the model showed 40% improvement in tasks requiring multi-stage reasoning compared to Opus 3.

In programming, the model can now edit entire source bases with a context of up to 200K tokens, understanding dependencies between files and maintaining style consistency. Developers who tested the template in preview report that it "understands the intention behind the code, not just the syntax".

Benchmark highlights

BenchmarkOpus 3Opus 4MelhoriaMMLU-Pro78.2%87.1%+11.4% HumanEval84.9%93.7%+10.4% MATH67.3%82.5%+22.6% ARC-AGI21.4%53.8%+151%

Availability

The template is available today in the Anthropic API and the web console. The price is $15 per million input tokens and $75 per million output tokens — aligned with front models of competitors.

Anthropic also announced that Claude Sonnet 4, more accessible version, will be released in the coming weeks with "90% of Opus 4's capacity to a third of the price".

Share:

Source: www.anthropic.com

Manu Ramalho

Written by

Manu Ramalho

Sou Manu Ramalho, publicitária com 16 anos de estrada conectando marcas e pessoas. Como fundadora da EME Marketing Digital, sempre busquei o marketing estratégico para gerar conexões autênticas. Aqui, mergulho na fronteira da inteligência artificial como analista de tendências. Meu foco é traduzir a complexidade de NLP, novos modelos de linguagem e papers acadêmicos para o mundo real, sempre com um olhar atento à regulamentação, ética e aos impactos sociais que essa tecnologia imprime na nossa sociedade.

Related articles