Anthropic Releases Claude Opus 4.7 With xhigh Effort Level
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Derya Unutmaz reports on X that Opus 4.7's graduate-level reasoning is now on par with GPT-5.4 and Gemini 3.1 Pro, calling it state-of-the-art for agentic coding and computer use
- Laki notes the model improves on code by +10% and image/graph understanding by +13% over the previous version
- Help Net Security uniquely highlights that the release includes automated cybersecurity safeguards — an angle most other coverage skips in favor of the coding story
- Nathan Lambert points out Opus 4.7 has a new tokenizer, making it a new base model rather than an incremental update from 4.6
- Haseeb and Matt Shumer flag that Anthropic is now showing benchmarks alongside a withheld 'Mythos' model, with Haseeb calling it 'the undemocratic era of AI' where the most intelligent model isn't publicly available
- Alex Albert highlights improved async work and instruction-following, more predictable effort levels for token control, no downscaling of high-res images, and 'noticeably more taste in UIs, slides, docs'
Why it matters: By shipping a new-base-model Claude (Nathan Lambert's new-tokenizer observation) with modest +10% coding gains while publicly benchmarking it against a withheld 'Mythos' frontier model, Anthropic has put the two-tier access structure on display. Developers can see the gap in the same chart, and the frontier model is no longer a rumor — it's a named, benchmarked product they're explicitly not getting.
Ask SkimNews
