If you saved yesterday’s live-blog playbook and you’re staring at your calendar wondering what to actually do this morning before tomorrow’s keynote — this is the companion read. The keynote starts at 9:30 a.m. Pacific tomorrow, May 6. Right now is the T-1 window where the next 24 hours of dev-press posts will be written.
Anthropic also has a quieter event today at 11 a.m. Eastern — The Briefing: Financial Services, virtual — and the pattern from previous Anthropic events is that vertical briefings 24 hours before a developer conference often preview what lands at the bigger event the next morning.
Here’s what to read before the keynote, and the 3 Sonnet 4.8 predictions that are now lockable given today’s signal.
What to Read This Morning
Before the keynote, four URLs are worth the bookmark:
- Anthropic’s blog feed — they post the morning of major announcements, often 30-90 minutes before the keynote starts. (anthropic.com/news)
- The Code with Claude SF agenda page — confirmed speakers Ami Vora (Head of Product), Boris Cherny (Head of Claude Code), Caitlin Colgrove, Martin DeVido, Thariq Shihipar, Nick Khami. The session list reveals what’s about to be announced. (claude.com/code-with-claude/san-francisco)
- The npm registry / Anthropic SDK changelog — Sonnet model identifiers sometimes surface in package metadata before the keynote.
- Vertex AI and Bedrock model lists — same logic. If a new model identifier is enabled before 9:30 a.m. Pacific, the keynote is a coronation, not a reveal.
The session list itself (Anthropic confirmed) covers three tracks:
- “What today’s models can do and where they’re headed”
- “Production-grade agents on the Claude Platform”
- “Claude Code at scale — long-horizon tasks, multi-repo work, parallel agents”
Each of these tracks has a feature waiting to be announced. Track 1 is the model-capability slot — historically where Sonnet upgrades have landed. Track 3 is where multi-repo / parallel-agent infra has been queued up since the Claude Code 2.1.126 Project Purge release on May 1.
The Briefing: Financial Services Today (11 a.m. Eastern)
Before tomorrow’s keynote, today’s FS-vertical event is worth a look — even if you’re not in financial services. (Anthropic event page)
The pattern: Anthropic uses vertical events to preview platform-level announcements 24-48 hours before the bigger conference. If the FS Briefing today demos something that looks like a new Claude Cowork skills bundle, a Claude × Excel integration, or a Mythos/Glasswing partner expansion, that’s almost certainly the same feature getting a wider announcement tomorrow.
Three things to specifically watch for at 11 a.m. ET:
- A Claude Cowork financial-services skills bundle. This would be a leading indicator for a broader Skills marketplace announcement at the Wednesday keynote.
- A Claude × Excel or Claude × Bloomberg integration. Either would preview a connector-pattern announcement at SF.
- A Mythos / Glasswing partner-tier expansion beyond JPMorgan as the existing launch FI partner. If Anthropic adds 1-2 more FS firms today, expect a wider Glasswing announcement Wednesday.
The post-event Anthropic blog post usually lands within 2 hours. Read both — that’s your one-day-out preview.
The 3 Sonnet 4.8 Predictions Now Lockable
Given T-1 signal — confirmed agenda, recent Anthropic blog cadence, the Sonnet 4.6 cycle behavior, and the npm leak referenced in AI Nexus Daily’s reporting on a Sonnet 4.8 source-code reference — three things are predictable enough to act on:
Prediction 1: Sonnet 4.8 lands tomorrow morning before the 9:30 a.m. Pacific keynote.
The signal: npm leaks, Vertex AI page references, model-list endpoint behavior, and the structure of the keynote agenda (Track 1 starts the day and historically opens with the model upgrade). The pattern matches the Sonnet 4.6 launch in November.
If you’re an eng lead, the operational consequence is: someone on your team should be running model-list checks at 7:00 a.m. Pacific tomorrow morning. Five endpoints to hit:
- Anthropic API model-list (
https://api.anthropic.com/v1/models) - Vertex AI Anthropic model page on Google Cloud
- AWS Bedrock model-list
- npm registry for
@anthropic-ai/sdkpatch versions - The Anthropic blog feed
Prediction 2: Pricing holds at $3 / $15 per million tokens.
Anthropic has held Sonnet pricing steady across 4.5 → 4.6 cycles. There’s no public signal of a pricing change in the run-up to 4.8. If pricing changes at all tomorrow, expect it on Opus tier, not Sonnet.
For your team’s prod-cost projections, plan for unchanged Sonnet pricing.
Prediction 3: At least one Anthropic Skills marketplace expansion lands at SF.
Boris Cherny is on the agenda; Skills are his lane. The Claude Cowork track (per session listing) is the natural venue for a Skills-marketplace announcement. The pattern from prior Anthropic events is that a marketplace announcement comes paired with 5-10 launch partners visible on the demo screen.
If you’ve been waiting to standardize on Skills versus building custom integrations, tomorrow is the day to wait through.
What This Means for You
If you’re an eng lead at a Claude-anchored shop: Block 7:00–8:00 a.m. Pacific tomorrow morning for someone on your team to run the 5 model-list checks. If Sonnet 4.8 is live before the keynote, you have 90 minutes to draft a hot-pin policy (“stay on 4.6 for prod for the first 24 hours”) and post it in the team channel before 9:30 a.m. The May 4 production-readiness audit playbook V2 shipped covers what to actually run.
If you’re a developer planning to watch the livestream: The keynote is at 9:30 a.m. Pacific. Set 3 calendar reminders — pre-keynote (7:00 a.m. PT, model-list check), mid-keynote (10:00 a.m. PT, screenshot any Skills marketplace screen), post-keynote (12:00 p.m. PT, read the Anthropic blog post that always lands within 2 hours).
If you’re a CTO at a non-AWS, non-GCP shop using Anthropic API direct: The model-pin question is the one to flag for your team. If Sonnet 4.8 ships tomorrow, you can probably default-route to it within a week — but your eval suite needs to run first. Don’t auto-pin.
If you’re a financial-services tech buyer: Today’s FS Briefing at 11 a.m. ET is more relevant for you than tomorrow’s developer conference. The leading-indicator features (Cowork bundle, Excel integration, Glasswing tier) are the ones to watch. The ROI math on each becomes available within 48 hours.
If you’re a journalist or analyst writing the keynote recap: The same-day blog from Anthropic is the cleanest source. It lands 30-90 minutes after the keynote and structures the announcement list exactly as Anthropic wants it framed. The agenda page session titles tell you the feature labels in advance.
What These Predictions Don’t Cover
KAIROS persistent agents / Undercover Mode could land tomorrow. The April 2 leak referenced KAIROS by name; it has been in the watchlist since then. If Anthropic announces it at SF, all the predictions above re-shuffle because KAIROS is a structural change to how agents persist across sessions, not just a model bump.
A Cowork mode GA + skills-marketplace expansion combo would also re-shuffle. Right now Cowork is in preview. A GA + marketplace combo is a bigger announcement than a Sonnet 4.8 upgrade, and it would dominate the keynote even if Sonnet 4.8 ships the same morning.
A major Mythos / Glasswing OT statement would re-shuffle. OT industry groups have been publicly frustrated about exclusion from the Glasswing partner tier. If Anthropic addresses this at SF, the developer-press coverage gets eaten by national-security press.
Sonnet 4.8 might not actually launch tomorrow. The npm leak is the strongest signal but it could refer to internal naming for a future release. If Sonnet 4.8 doesn’t land tomorrow, the production-readiness audit playbook moves from “run at 7 a.m.” to “run when the launch announcement actually happens.” Don’t pre-commit infra changes on a model that hasn’t shipped yet.
Anthropic could announce something not in the npm leak or agenda. The 3 things this post calls out are the predictable surprises. Anthropic may bring something not predicted at all — a partnership, a vertical bundle, a sandbox change. Those are by definition not lockable in advance.
The Bottom Line
The hour before tomorrow morning’s keynote is more useful than the hour after. If you’ve already saved yesterday’s live-blog playbook, this is the operational layer that makes it actionable: read the FS Briefing today, run model-list checks at 7 a.m. Pacific tomorrow, watch for Skills marketplace and Cowork GA announcements. Treat the 3 predictions above as locks worth acting on; treat KAIROS and Glasswing-OT as the surprises.
For the operational side of running Claude in production at scale, our Claude Code Mastery course is built for the eng leads who’ll be making the model-pin decisions tomorrow morning. Claude Code Routines Mastery covers the long-horizon, multi-repo workflows that Track 3 of the keynote will be all about.
Set the alarm. Open the model-list endpoints. See you at 9:30 Pacific.
Sources
- Code with Claude San Francisco — May 6, 2026 (event page)
- Code with Claude — Anthropic’s Developer Conference
- Code with Claude comes to San Francisco, London, and Tokyo — Anthropic blog
- The Briefing: Financial Services — Anthropic event page
- Anthropic Source Code Leak Reveals Claude Sonnet 4.8 — AI Nexus Daily
- Claude Sonnet 4.8 Release Date Speculation — NxCode
- Anthropic Events
