OpenAIAnthropicGooglexAIPerplexity
Nova3 / current state

The state of the frontier, right now.

One read. Refreshed weekly. Drawn from the 40 most material moves across 5 frontier labs in the last 7 days.

By Mark JonesUpdated
Window: Jul 11, 2026 to Jul 18, 2026·Generated Jul 18, 2026, 4:34 PM (CT)
This week's read

AI shifted from chat answers into connected task surfaces with tighter context, cost, and controls

The week's common thread is operationalization: memory stays current, context gets compacted, tasks move through connected apps, and voice, media, batch, browser, desktop, and WebSocket paths become normal surfaces. The operator move is to treat model choice as plumbing, then wire agents around context freshness, app permissions, request-level cost, and injection controls.

Quiet this window

No material moves from xAI in the last 7 days. Tracked, just not active.

Sources informing this read (40)
  1. Jul 18, 2026
    AnthropicAnthropic model card shows no readable text change in this diff.
  2. Jul 18, 2026
    AnthropicAnthropic pricing page removed the Claude plan recommendation questionnaire content.
  3. Jul 18, 2026
    OpenAIChatGPT scheduled tasks gained a sidebar page for viewing, managing, and editing active tasks.
  4. Jul 18, 2026
    GoogleGemini pricing page removed captured pricing-plan copy and Interactions API banner.
  5. Jul 18, 2026
    AnthropicAnthropic lists Claude Fable 5 as its most capable widely released model.
  6. Jul 18, 2026
    AnthropicAnthropic pricing page removed a five-question plan recommendation flow.
  7. Jul 17, 2026
    AnthropicAnthropic pricing removed the Claude plan quiz recommending Team or Enterprise self-serve.
  8. Jul 17, 2026
    OpenAIOpenAI pricing page removed Business and Enterprise plan copy.
  9. Jul 17, 2026
    AnthropicAnthropic retired Claude Sonnet 4 and Claude Opus 4, and requests to them now return errors.
  10. Jul 17, 2026
    OpenAIOpenAI pricing page removed API pricing table content for listed GPT-5.6 models
  11. Jul 17, 2026
    AnthropicAnthropic removed model card text for Claude Fable 5 and Claude Mythos 5.
  12. Jul 17, 2026
    AnthropicAnthropic removed the monitored model card page’s model overview and comparison content.
  13. Jul 17, 2026
    AnthropicAnthropic pricing page removed team sizing and contract-fit questionnaire content
  14. Jul 17, 2026
    GoogleGoogle pricing snapshot removed Gemini API pricing navigation and availability note.
  15. Jul 17, 2026
    OpenAIOpenAI updated the ChatGPT desktop app for Chat and Work selection, search, Projects, and cross-device Work.
  16. Jul 17, 2026
    OpenAIChatGPT desktop added a ChatGPT and Codex switcher, unified Recents, Projects, and cross-device Cloud Work sync.
  17. Jul 16, 2026
    GoogleGoogle pricing diff removed Gemini API model navigation and Interactions API note
  18. Jul 16, 2026
    AnthropicMid-conversation system messages are available on listed Claude models across API, Bedrock, and Google Cloud without beta headers.
  19. Jul 16, 2026
    AnthropicAnthropic removed Fable 5 and Mythos 5 guidance from its modelcard page.
  20. Jul 16, 2026
    AnthropicAnthropic pricing no longer shows the Claude plan recommendation questionnaire.
  21. Jul 15, 2026
    AnthropicClaude Enterprise admins can now manage organization people through the beta Admin API.
  22. Jul 15, 2026
    AnthropicAnthropic pricing page removed its team-plan questionnaire and self-serve Enterprise recommendation content
  23. Jul 15, 2026
    OpenAIOpenAI increased ChatGPT custom instructions to 5,000 characters for Plus, Pro, Enterprise, Business, and Education users.
  24. Jul 15, 2026
    OpenAIOpenAI model card page removed API docs navigation text.
  25. Jul 15, 2026
    GoogleGoogle removed visible Gemini API pricing page content from the monitored output
  26. Jul 15, 2026
    AnthropicAnthropic removed model overview text covering Fable 5, Mythos 5, and availability.
  27. Jul 14, 2026
    AnthropicAnthropic removed modelcard copy for Claude Fable 5 and Claude Mythos 5.
  28. Jul 14, 2026
    OpenAIOpenAI retired GPT-5.2 models in ChatGPT; existing conversations continue on corresponding GPT-5.5 models.
  29. Jul 14, 2026
    AnthropicClaude Managed Agents webhooks, multiagent orchestration, and self-hosted sandboxes are now available on Claude Platform on AWS.
  30. Jul 14, 2026
    AnthropicMultiagent orchestration, Outcomes, vault credential refresh, and webhooks are now supported for Claude Managed Agents.
  31. Jul 14, 2026
    PerplexityPerplexity Computer now remembers past work, responds faster, switches models, publishes websites, and researches private companies.
  32. Jul 14, 2026
    PerplexityComputer is available in Microsoft 365 apps with clearer thread context, visible sources, and usage analytics.
  33. Jul 14, 2026
    PerplexityComputer is faster, more powerful, and more connected to enterprise with improved models, Teams, and Spaces skills.
  34. Jul 14, 2026
    AnthropicDreams research preview now supports Claude Fable 5 and Claude Sonnet 5.
  35. Jul 14, 2026
    PerplexityPerplexity brought Deep Research to Computer with command access, forking, inline actions, analytics APIs, and credit limits.
  36. Jul 14, 2026
    PerplexityPerplexity made Personal Computer available for everyone on Mac and ready for all finance professionals.
  37. Jul 14, 2026
    PerplexityPersonal Computer on Mac shipped, with Computer updates for control, personal context, and a prebuilt Personal CFO.
  38. Jul 14, 2026
    OpenAIOpenAI removed ChatGPT Business and Enterprise pricing copy from the pricing page
  39. Jul 14, 2026
    AnthropicAnthropic removed modelcard content naming Claude Fable 5 and Mythos 5 availability.
  40. Jul 14, 2026
    AnthropicAnthropic pricing removed the five-question Claude plan recommendation flow from the captured page.
FAQ
What is the current state of frontier AI?+

Nova3 publishes a single weekly read synthesizing the most material moves across OpenAI, Anthropic, Google, xAI, and Perplexity in the last 7 days. The full read sits at the top of this page, with every source listed below it.

How often is this updated?+

The pool of frontier moves updates continuously as a Nova3 agent detects new launches. The weekly read is regenerated at least once per 7-day window. The header shows the current window and the last generation timestamp.

Which labs does Nova3 track?+

OpenAI, Anthropic, Google (DeepMind and the Gemini surface), xAI, and Perplexity. Each lab carries an entity link to its Wikidata page so machine readers can resolve the reference unambiguously.

How does Nova3 grade severity?+

Three tiers: critical (changes how something gets built), material (changes who can build it or at what cost), minor (notable but non-structural). A governance layer blocks unsourced claims, so every graded entry carries a primary source link.

Can I get this as raw markdown for an agent?+

Yes. /current-state.md mirrors the same payload as plain markdown for ingestion by LLMs and agentic search clients. /llms.txt and /llms-full.txt expose the broader site corpus.

This is the world we build in. It moves this fast. We move with it.

Bring us your project →

Prefer raw? Read the markdown · See the feed