Report: Claude Opus 5 surfaces in Vertex catalog; launch imminent
Leaked specs point to a 1M-token window and an extra-high-effort mode; Polymarket briefly put a Jul 24 debut at 88%. Also: OpenAI Presence has no self-serve.
Copy markdown
The tell: a model string in Google's Vertex catalog
A Claude Opus 5 entry briefly appeared in a coding tool's model picker before being pulled, then a matching string surfaced on Google's Vertex quotas catalog — the same pre-launch sequence that reportedly preceded Sonnet 5. Anthropic has said nothing official yet.
Leaked spec: 1M context and an 'extra-high-effort' mode
The vanished entry reportedly listed a one-million-token context window, per-turn controls, safety fallbacks, and an 'extra-high-effort' reasoning setting — a dial past Opus 4.8's current ceiling for the hardest agentic runs.
What it means for your 4.8 pipelines
Opus 5 would reportedly land across Claude apps, Claude Code, the Claude Platform, and cloud partners, likely replacing Opus 4.8 rather than running beside it — so pin your model IDs now. It's been 56 days since 4.8, and Polymarket briefly put a Jul 24 debut at 88%.
OpenAI ships Presence — but there's no API to touch
OpenAI launched Presence, an enterprise platform for realtime voice and chat agents that resolves 75% of its own English phone support. The catch for builders: no self-serve and no SDK — deployments run through Forward Deployed Engineers only.