Claude 4.5 Haiku (thinking)

Claude Sonnet 4.5 is Anthropic's most advanced Sonnet model, optimized for AI agents and coding workflows. It delivers superior performance on coding benchmarks and introduces powerful agentic capabilities such as tool orchestration and speculative parallel execution. This model is suited for multi-context and long-horizon workflows, capable of operating autonomously for many hours.

Claude Opus 4.6 (Non-thinking) — the default high-effort version — is already among the top models for reasoning capability, while the low-effort version thinks less but shows little difference in output. Despite its high cost, it offers multimodal input processing including text and images, while generating quality text output. A key highlight is its context window that can scale up to 1 million tokens, enabling the processing of large volumes of information.

Claude Opus 4.6 (Thinking) is one of the models primarily focused on Adaptive Thinking. Despite being expensive, slow, and verbose, it excels with adaptive reasoning capabilities. This model supports text and image input, then outputs text.

Claude Opus 4.7 (Non-reasoning, High Effort) is one of Anthropic's best models. Despite reduced reasoning, with high effort enabled the model remains very powerful. It supports text and image input, with text output. However, this model is quite expensive, slower than average, and tends to overthink.

Claude Opus 4.7 (Adaptive Reasoning, Max Effort) is one of the leading intelligence models, supporting text and image input while outputting text. It features adaptive reasoning capabilities and is designed for tasks requiring maximum effort. This model is very detailed in its responses.

Claude Opus 4.8 (Adaptive Reasoning, Max Effort) is one of the top models in terms of intelligence, supporting text and image input, and text output. This model excels at complex tasks and is slower than average compared to other models in the same segment. It is also quite verbose in its responses, so Anthropic has added a fast feature for this model, notably with the price remaining unchanged at $6.25/1M input tokens (cache $0.50) and $25/1M output tokens.