Act on this
Future Claude models may embed text watermarks under EU AI Act
Check Anthropic's documentation on watermarking before relying on output provenance for future Claude models.
- Signal
- 64
- Facts
- 69
- Hype
- 7
Claude model release
Topic
19 cards
Act on this
Claude Code scheduled tasks revert model edits while the desktop app runs
Close the desktop app before editing model settings, or run a skill that switches tasks to Sonnet 5.5.
- Signal
- 80
- Facts
- 64
- Hype
- 1
Brief
Crypto outlet says Opus 5.5 tops a new AI index at 61.7
The report could not be checked, since the source page is blocked and no other outlet confirmed the score. Treat the ranking as unverified.
- Signal
- 29
- Facts
- 42
- Hype
- 29
Brief
LiveNerf tracker finds no Opus 5.5 nerf over 29 days
A community tracker reports 29 days of capability data with no drop beyond its 6.6 point threshold. It is self-reported and unfunded.
- Signal
- 32
- Facts
- 89
- Hype
- 20
Brief
TechCrunch frames Claude Duet as Anthropic's push into proactive agents
The analysis places Duet among proactive business agents, waitlisted for US Pro and Max users. It is built on Opus 5.5 with 6,000 connectors and approval gates.
- Signal
- 44
- Facts
- 52
- Hype
- 28
Brief
Japanese roundup covers Duet preview, OpenAI on Google Cloud and a PHP flaw
The note.com digest says Claude Duet is a research preview for US paid accounts, not yet open to Japan. It also covers a critical PHP vulnerability.
- Signal
- 68
- Facts
- 65
- Hype
- 46
Hype watch
CZ jokes that Claude Duet now controls us
The reaction is commentary on Anthropic's Claude Duet, a background agent that asks before acting. It adds no new technical details.
- Signal
- 27
- Facts
- 34
- Hype
- 42
Act on this
Claude API version 2026-10-21 removes temperature, top_p and top_k
Check code that sets temperature, top_p or top_k before upgrading, since the new version drops them from the Messages API.
- Signal
- 97
- Facts
- 59
- Hype
- 2
Brief
Amodei says Anthropic held back a Mythos-class model for four months
Dario Amodei said on X on October 20 that the model was delayed to finish cyber safeguards. He was rebutting a critique from Nathan Calvin.
- Signal
- 31
- Facts
- 67
- Hype
- 10
Act on this
Opus 5.5 Fast mode is out of research preview
Enable it with speed: fast in the API or /fast in Claude Code for up to 2.5x output speed at a higher price.
- Signal
- 92
- Facts
- 83
- Hype
- 2
Brief
Claude Duet is Anthropic's proactive always-on personal agent
A research preview for US Pro and Max subscribers via waitlist, powered by Opus 5.5. Team and Enterprise plans are excluded.
- Signal
- 43
- Facts
- 60
- Hype
- 42
Act on this
Claude Security adds continuous scanning and fix tracking in public beta
Enterprise teams can try the beta, capped at eight repos per project, with $25,000 monthly credits through December 31.
- Signal
- 87
- Facts
- 68
- Hype
- 18
Brief
Developer asks how Argon compares to Opus 5.5 for game art judgment
A question post with no benchmarks or vendor facts. The Argon comparison is unverified.
- Signal
- 18
- Facts
- 9
- Hype
- 10
Brief
Reddit user reports Opus 5 losing language precision
A single anecdote about arithmetic slips and clarifying questions. There is no official change notice, version or date.
- Signal
- 41
- Facts
- 6
- Hype
- 0
Brief
Report: Opus 5.5 quietly switches to Opus 4.8 for a security safeguard
An MSN report says Anthropic's own testing calls the switch a weak point. The cached text is thin and the switch claims are unverified.
- Signal
- 74
- Facts
- 30
- Hype
- 18
Brief
Report: Anthropic is building a model that infers user intent
Per The Information, the model would infer user goals and find root-cause fixes. Name, specs and date are unknown, and Anthropic has not announced it.
- Signal
- 27
- Facts
- 33
- Hype
- 23
Act on this
Claude Haiku 5.5 launches at $0.10 input per million tokens
Recheck your Haiku budget, since the tokenizer runs about 1.25x heavier than Haiku 4.5.
- Signal
- 97
- Facts
- 66
- Hype
- 14
Hype watch
Reddit reports Claude Fable 5.5 from an Azure listing
Based on a brief Azure catalog listing and LiteLLM routing entries. The endpoint is not live and the spec is unofficial, so do not hardcode it.
- Signal
- 34
- Facts
- 54
- Hype
- 65
Brief
Reddit users ask when Opus became the default over Sonnet
One user noticed weaker Sonnet answers on reasoning tasks. No official default change is cited in the thread.
- Signal
- 41
- Facts
- 5
- Hype
- 2