Everything published on Agent Upstart, newest first
This page lists everything published on Agent Upstart, newest first. The newest work sits at the top, and every entry names its section.
- Claude Code hardens its sandbox ahead of Nov 10Developer · changelog read · Claude Code sandbox
- Claude Design closes December 14: export nowClaude user · shutdown read · Claude Design closing
- Antigravity CLI 1.5.3: subagent models, sandbox fixDeveloper · changelog read · Antigravity CLI 1.5.3
- GPT-5.5-Cyber preview ends February 1, 2027Developer · deprecation read · GPT-5.5-Cyber sunset
- Run Muse Spark in another harness via the Model APIDeveloper · docs read · Muse Spark Model API
- Haiku 5.5 is now the default HaikuDeveloper · model read · Haiku 5.5
- PACT-Check: benchmark gains without information flowBuilder · benchmark read · PACT-Check
- Scientist World: 1,000 executable research tasksBuilder · benchmark read · Scientist World
- Kilo's November 12 cutoff: what actually endsTeam lead · cutoff read · Kilo November 12
- Hermes Agent 0.23.0: paid MCP spend limitsDeveloper · changelog read · Hermes Agent 0.23.0
- Anthropic's no-distillation rule lands November 12Developer · policy read · Anthropic distillation ban
- Mistral Vibe 3.1.0: see what fills the contextDeveloper · changelog read · Mistral Vibe 3.1.0
- Auggie 0.37.0: first release after the takeoverDeveloper · ownership read · Auggie 0.37.0
- Codex Guardian v3: the evidence pipeline in 0.168Developer · feature read · Codex Guardian v3
- MCP's July spec killed sessions — tooling may not knowDeveloper · protocol read · MCP stateless spec
- TwIL-8B: a local model for math and codingDeveloper · model read · TwIL-8B
- Changelog: Agent UpstartAbout · changelog · what changed
- How we check thingsAbout · method · how pages are checked
- What we believeAbout · positions · what this site holds
- Snorkel reward-code reproductionExperiment · reproduction · Snorkel reward code
- Stroq attack suite: 36 scenariosExperiment · self-test · Stroq attack suite
- Aider /test passes arguments throughExperiment · behaviour test · aider /test
- ChatGPT and Codex share one token budgetDeveloper · change read · shared token budget
- Claude taught small models to deceive monitorsBuilder · evaluation read · subtle misalignment
- Claude Code sent your email to websitesDeveloper · privacy fix · Claude Code email leak
- OpenRouter Auto Code: Opus quality, 74% cheaper?Developer · launch read · OpenRouter Auto Code
- Claude reads your bank statements now — in the USClaude user · launch read · Personal Finance beta
- Google ADK 2.14: A2A graduates, Teams joins inDeveloper · release read · Google ADK 2.14.0
- Amp Spaces: orchestrator over parallel threadsDeveloper · launch read · Amp Spaces
- Inkling 2 Small: 17B active, 594B totalBuilder · launch read · Inkling 2 Small