24 July 2026

• Default model shake-up: OpenAI dropped GPT-4.5 from ChatGPT in favor of GPT-5.5 Instant as default; Google's Gemini 3.5 Flash now powers AI Mode and Search agents worldwide.

• Shift from tools to agents: Google is building Gemini into a multi-agent, task-oriented layer within Search; Canva and Adobe are embedding deeper conversational AI into their creative/design workflows.

• Broader model upgrade landscape: New releases across the board — GPT-5.6 family, Claude Sonnet 5, various Gemini Flash variants, Grok 4.5, plus maturing open-weight models — are collectively improving reasoning, speed, and workflow capability for research-heavy use cases.

• What this means for genealogy work:

  • Reasoning/agentic models are now capable of handling complex, multi-step projects — e.g., reconciling conflicting identities, building out timelines.
  • Fast, lightweight models are well-suited to high-volume tasks — transcription, translation, extraction, and batch cleanup across many ancestor records.

• Practical takeaway: More than 25 ready-to-use micro-workflows are available now, including:

  • GPT-5.6 Sol → ranked research/search plans
  • Claude Sonnet 5 → proof summaries
  • Gemini Flash → handwriting transcription
  • Grok 4.5 → structured JSON extraction
  • Local open-weight models → private/offline transcription

Plug-and-play micro-workflows

1. Use GPT-5.6 Sol to convert a research log into a ranked to-do list: Paste a month of notes and ask for “top 10 next searches, ordered by evidence value.”openai

2. Use GPT-5.6 Terra for a fast source inventory: Feed it a family group sheet and have it list missing census, probate, deed, military, and land records.openai+1

3. Use GPT-5.6 Luna for batch annotation: Ask it to tag 50 file names or citation snippets with surname, locality, and record type.openai+1

4. Use ChatGPT Voice with GPT-Live-1 for live brainstorming: Dictate a brick wall problem and have it help you outline hypotheses and next record groups.openai

5. Use GPT-5.6 max reasoning for conflict resolution: Give it two competing birth dates and ask for a source-by-source conflict analysis.openai+1

6. Use GPT-5.6 ultra mode for ancestor packet assembly: Break a large job into subasks such as timeline, locality summary, FAN-club analysis, and source gap list.openai+1

7. Use Claude Sonnet 5 to draft a proof summary: Provide extracts from deeds, probate, and census records and ask for a concise narrative with cited claim categories.anthropic

8. Use Claude Sonnet 5 to clean up a research report: Have it turn rough notes into a polished “findings, conflicts, next steps” memo.anthropic

9. Use Claude’s enterprise controls for client work: If you manage sensitive client files, separate organizational access and admin permissions before sharing research sets.anthropic+1

10. Use Gemini 3.6 Flash for handwriting triage: Upload image snippets from ledgers or court minutes and ask for a quick first-pass transcription.blog

11. Use Gemini 3.5 Flash-Lite for translation sprints: Push batches of foreign-language church, civil, or immigration records through a lightweight translation pass.blog

12. Use Gemini 3.5 Flash Cyber for risky-text review: When handling OCR from messy scans or web-scraped text, ask it to flag suspicious strings, malformed dates, and corrupted place names.blog

13. Use Gemini Live Translate for multilingual correspondence: Translate cousin letters or archive replies in near-real time during a research session.blog

14. Use Gemini Search/AI Mode for locality context: Ask for a county-history briefing, then compare it against your own deeds and probate timeline.blog

15. Use Grok 4.5 on the xAI API for scripted extraction: Feed it a stack of record summaries and ask for JSON output with names, dates, places, and record types.x

16. Use Grok 4.5 for agentic research batching: Split a project into subagents for census, military, land, and probate leads.x

17. Use Grok 4.5 for quick research ideation: Ask for alternate surname spellings, locality variants, and neighbor clusters before you search.x

18. Use an open-weight model for private transcription: Run a local model on sensitive family notes or client material when you do not want cloud processing.tech-insider+1

19. Use an open-weight model for Zotero cleanup: Have it standardize tags, folder names, and note templates across a local corpus.tidqom+1

20. Use an open-weight model for repetitive extraction: Batch a folder of OCR’d probate pages and pull out names, heirs, dates, and witnesses.tech-insider+1

21. Use GPT-5.6 Terra or Claude Sonnet 5 for FAN-club mapping: Ask for associates, neighbors, and repeat witnesses across multiple counties.openai+1

22. Use Gemini Flash or GPT-5.6 Luna for census comparison: Compare two household transcriptions side by side and flag likely enumeration errors.openai+1

23. Use Claude Sonnet 5 for a narrative from structured data: Give it a timeline table and ask for a readable ancestor sketch without inventing facts.anthropic

24. Use GPT-5.6 Sol for a “research audit”: Ask it to identify unsupported claims in a draft family biography before publication.openai

25. Use any of the new fast models for source triage in Oklahoma and Five Tribes work: Feed in allotment, probate, land, or tribal record excerpts and ask for locality, date range, and relationship clues.blog+2

A simple example: take one ancestor’s probate packet, run OCR cleanup in Gemini Flash-Lite, then ask GPT-5.6 Sol to build a chronology and Claude Sonnet 5 to draft the narrative paragraph. That division of labor is the fastest way to get both speed and quality without overloading one model with every task.