● AI RADAR BETA | Vol 01·Issue 008·Saturday, Jul 4 PT |
|
● LIVE INGEST 09:33 PT·JUN 27 TO JUL 3, 2026 i wake up, there is a busier week of AI updates.One issue, every drop linked, clustered by what the news meant. | 33 DROPS 3 TOOLS 5 THEMES |
|
↺ ░▒▓▒░ ↻
/\_/\
( ◕ω◕ )
> ^ <
i just wake up The week in themes.Jun 27 to Jul 3 · 33 drops |
|
Sonnet 5 arrives, priced to spread.Anthropic shipped Sonnet 5 as the new default across Claude and made it the headline benchmark story in Cursor. The pitch is Opus-adjacent quality at Sonnet prices, with agentic finishing behavior called out as the differentiator. READ DROPS ↓ |
Fable 5 returns with new classifiers.Commerce lifted export controls, Anthropic redeployed Fable 5 globally with tighter cybersecurity classifiers, and the company used the moment to float an industry framework for scoring jailbreak severity. The trade is a slightly higher false positive rate on harmless coding requests for continued access to the model. READ DROPS ↓ |
Claude Science, GeneBench-Pro, a life sciences hackathon.Anthropic and OpenAI both pointed their coding agents at wet lab and computational biology work in the same week. Claude Science ships as an application layer; GeneBench-Pro tries to measure whether agents can actually navigate messy biological data. READ DROPS ↓ |
Cursor on iOS, Claude on Azure, Codex at HP.The week's other spine was where these agents actually run. Cursor put always-on agents on the phone, Claude went generally available inside Microsoft Foundry, and OpenAI turned an HP rollout into a case study. READ DROPS ↓ |
UI cleanup, an 18-year-old bug.OpenAI spent the week on Codex ergonomics: paste behavior, navigation, a shortcuts refresh, and one long engineering post on a bug that had been quietly crashing infrastructure for eighteen years. A Steipete talk framed the direction: engineers set direction, agents do the work. READ DROPS ↓ |
|
THE FULL DIGEST, WITH SOURCES Every drop, themed and linked. | 24 linked drops
6 active days
3 tools tracked
|
|
01 MODEL REFRESH · 5 DROPS Sonnet 5 arrives, priced to spread.Anthropic shipped Sonnet 5 as the new default across Claude and made it the headline benchmark story in Cursor. The pitch is Opus-adjacent quality at Sonnet prices, with agentic finishing behavior called out as the differentiator. | CLAUDE | Sonnet 5. Frontier performance across coding, agents, and professional work. | TUE · RELEASE | → |
|
| CLAUDE | Sonnet 5. Default on Free and Pro, live across Max, Team, and Enterprise with introductory pricing through August 31. | TUE · 11:00 PT | → |
|
| CLAUDE | Sonnet 5. A substantial step over 4.6, close to Opus 4.8 at lower prices. | TUE · 11:00 PT | → |
|
| CLAUDE | Sonnet 5. Early partners say it finishes complex tasks and checks its own output. | TUE · 11:00 PT | → |
|
| CURSOR | Sonnet 5 in Cursor. 57 percent on CursorBench, up from 49 for Sonnet 4.6. | TUE · 11:13 PT | → |
|
|
|
02 TRUST AND SAFETY · 6 DROPS Fable 5 returns with new classifiers.Commerce lifted export controls, Anthropic redeployed Fable 5 globally with tighter cybersecurity classifiers, and the company used the moment to float an industry framework for scoring jailbreak severity. The trade is a slightly higher false positive rate on harmless coding requests for continued access to the model. | CLAUDE | Commerce Department lifted export controls on Claude Fable 5 and Mythos 5. | TUE · 16:52 PT | → |
|
| CLAUDE | Fable 5 redeployed globally with new classifiers targeting cybersecurity misuse. | TUE · 20:42 PT | → |
|
| CLAUDE | Redeploying Fable 5. Anthropic proposed a jailbreak severity framework with Amazon, Microsoft, Google, and Glasswing partners. | TUE · SAFETY | → |
|
| CLAUDE | Cybersecurity safeguards updated after conversations with the US government. | WED · 12:31 PT | → |
|
| CLAUDE | Paid plans access Fable 5 for up to 50 percent of weekly usage through July 7. | WED · 12:31 PT | → |
|
| CURSOR | Fable 5 back in Cursor. Leads CursorBench at the highest cost per task. | WED · 12:33 PT | → |
|
|
|
03 SCIENCE · 3 DROPS Claude Science, GeneBench-Pro, a life sciences hackathon.Anthropic and OpenAI both pointed their coding agents at wet lab and computational biology work in the same week. Claude Science ships as an application layer; GeneBench-Pro tries to measure whether agents can actually navigate messy biological data. | CLAUDE | Claude Science. Auditable artifacts, on-demand environments, and 60-plus scientific databases. | TUE · RELEASE | → |
|
| CODEX | GeneBench-Pro. A research-level benchmark for agents navigating biological data and analysis paths. | TUE · RESEARCH | → |
|
| CLAUDE | Built with Claude: Life Sciences. Virtual hackathon with the Gladstone Institutes and 100k in credits. | THU · 07:00 PT | → |
|
|
|
04 DISTRIBUTION · 5 DROPS Cursor on iOS, Claude on Azure, Codex at HP.The week's other spine was where these agents actually run. Cursor put always-on agents on the phone, Claude went generally available inside Microsoft Foundry, and OpenAI turned an HP rollout into a case study. | CURSOR | Cursor for iOS. Launch always-on cloud agents or remote-control agents on your desktop. | MON · 10:05 PT | → |
|
| CURSOR | Live Activities in Cursor. Notifications when agents finish, mobile diff review before merging PRs. | MON · 10:05 PT | → |
|
| CLAUDE | Claude in Microsoft Foundry generally available on Azure, with Opus 4.8 and Haiku 4.5. | MON · 10:56 PT | → |
|
| CURSOR | Cursor team marketplaces expand to team MCPs and organization groups. | TUE · PRODUCT | → |
|
| CODEX | HP partnership. OpenAI's writeup of scaling early AI wins across the enterprise. | SUN · COMPANY | → |
|
|
|
05 CODEX POLISH · 5 DROPS UI cleanup, an 18-year-old bug.OpenAI spent the week on Codex ergonomics: paste behavior, navigation, a shortcuts refresh, and one long engineering post on a bug that had been quietly crashing infrastructure for eighteen years. A Steipete talk framed the direction: engineers set direction, agents do the work. | CODEX | Codex threads. New navigation rail previews nearby turns on hover. | SAT · 10:31 PT | → |
|
| CODEX | Codex paste. Markdown into Slack preserves bullets, bold, code, and links. | SAT · 15:07 PT | → |
|
| CODEX | Codex shortcuts get an upgrade on July 15. | MON · 10:00 PT | → |
|
| CODEX | Core dump epidemiology. Fixing an 18-year-old open source bug in data infrastructure. | TUE · ENGINEERING | → |
|
| CODEX | Steipete at AI Engineer. Engineering shifts to direction, review, and system design as agents take longer work. | TUE · 12:15 PT | → |
|
|
|
i wake up, there is another one. TOOLS TRACKED CODEX · 9 CLAUDE · 18 CURSOR · 6
Built by Jane You with Claude Code & Design · Maintained by RaeyaBot · Daily ingest 5am PT read in browser · airadarapp.com · unsubscribe |
|
|