[KongchangAI]
ModelOpus 4 / Opus 4.5 / Opus 4.8

Opus 4.7

Opus 4.7 is an iteration of Anthropic's Opus series of large language models within the Claude model family. The model exhibits strong stability in long-horizon autonomous task execution, supports complex multi-step reasoning, and enables automation of software development workflows. It is designed for high-complexity tasks requiring sustained agentic behavior and represents one of Anthropic's models optimized for high-performance autonomous agent applications.

Timeline (last 90 days)

Oct 8

Opus 4.7 was released in April 2026, has a context window of one million tokens, and costs approximately $5 per million input tokens.

Unverified50%
Oct 7

Opus 4.7 can directly capture native-resolution screenshots and return 1:1 pixel coordinates, supporting up to 1440p.

Unverified50%
Oct 7

On the OS World benchmark, less than 12 months ago Claude scored below 50%, whereas Opus 4.7 has reportedly reached 78%.

Unverified50%
Oct 7

For ultra-high resolutions such as 4K, local downsampling is still recommended, while for resolutions up to 1440p, trying different resolutions and image formats (JPEG/PNG/WebP) is encouraged.

Unverified50%
Oct 1

Opus 4.7+ 和 Fable 模型现在在 Bedrock、Vertex、Foundry 以及 Claude apps 网关上默认使用100万(1M)上下文窗口,不再需要 [1m] 后缀

Unverified50%
Sep 29

Sessions behind a custom `ANTHROPIC_BASE_URL` will now use models supporting a 1-million-token context window (Opus 4.7+, Sonnet 5+, Fable); if the gateway truncates at 200K, `/autocompact 200k` can be run as a workaround.

Unverified50%
Sep 29

Versions Opus 4.7, 4.8, and 5.0 exhibited a decline in conversational quality, a heavier AI-like feel, and deteriorating task execution performance.

Unverified50%
Sep 23

The user stated that several versions starting from Opus 4.7 were unbearable.

Expired50%

All Facts (8)

Source Articles