As of this morning there are fresh roundups and documentation updates that clarify how to use the very latest models and features that landed earlier this week and last week.
While it's too early to see genealogy related work using An tropics' latest release - Claude Fable - a special note about it is included below.
A. Named releases & features (last few days)
These are releases, docs, or roundups updated in roughly the last 72 hours that matter for day‑to‑day genealogy work.
Perplexity – Multi‑Provider Model Comparison (updated June 10, 2026): New cookbook page shows how to send the same genealogy prompt to multiple top models (GPT‑5.4, Claude Sonnet 4.6, Gemini 3.1 Flash‑Lite, Grok 4.20, Perplexity Sonar) from one command‑line tool, making it far easier to A/B test responses on your real research questions.[docs.perplexity]
OpenAI – GPT‑5.5 and GPT‑5.5 Instant highlighted as June 2026 defaults: Current guidance now positions GPT‑5.5 (and its cheaper Instant variant) as the main “daily workhorse,” with strong reasoning, multimodal support, and good value, which affects which model you should pick inside tools that let you choose.[felloai]
Anthropic – Claude Sonnet 4.6 as new default: Sonnet 4.6 is documented as the default Claude model, bringing improved long‑context reasoning and “computer use” skills (navigating interfaces and files) at a lower cost than previous high‑end models.
(see below for note on Claude Fable)
Google – Gemini 3.1 Pro and 3.1 Flash‑Lite guidance (June 2026): Google’s newest documentation and comparisons emphasize Gemini 3.1 Pro for reasoning and multimodal understanding and 3.1 Flash‑Lite as a light‑weight, cheaper option; many front‑ends (including Perplexity API) now list these by name.[docs.perplexity]
xAI – Grok 4.2 / 4.20 multi‑agent architecture: Grok 4.2’s multi‑agent design (multiple specialized agents debating before answering) and the 4.20 configuration exposed in Perplexity’s cookbook give genealogists access to another style of cautious, debate‑based reasoning.[marketingprofs]
Perplexity – Personal Computer & computer‑use agents (rolling out this week): Perplexity’s “Personal Computer” feature, highlighted again in recent news, runs a 24/7 agent on your Mac that can orchestrate your browser, local files, and apps, useful for automating repetitive genealogy tasks like downloading images and filing them.[youtube]
Best‑model guidance for June 2026 (meta‑release): A June 10th analysis ranks GPT‑5.5, Claude Opus/Sonnet 4.6, and Gemini 3.x as today’s “top tier” for everyday work, which is effectively a cross‑vendor recommendation list you can adopt when choosing models inside generic AI tools you already use.[felloai]
On the genealogy‑specific side, no brand‑new platform tools appear to have launched in the last 72 hours, but FamilySearch’s February 2026 AI overview (still current) is being widely cited this week as the baseline for how mainstream genealogy platforms are integrating AI (AI‑indexed records, AI research assistants, etc.).[familysearch]
B. Implications for genealogists this week
The big story this week is not “one brand‑new genealogy AI,” but rather convergence on a short list of very strong general models and clearer guidance on which to use for what. For you, that means you can confidently standardize on 2–3 “go‑to” models across tools instead of constantly experimenting. In practice, that usually looks like GPT‑5.x or Claude Sonnet 4.6 for careful reasoning and writing, and a Gemini 3.1‑family model when images or handwriting are involved.[felloai]
Second, the Perplexity model‑comparison cookbook is a quiet but important upgrade for methodology‑minded researchers. It lets you send the exact same research prompt (for example, a locality survey for “Oklahoma Territory probate 1890–1907”) to several models at once and compare outputs side‑by‑side, which is ideal for teaching, documenting workflows, and testing reliability before you trust an AI‑assisted method in a client project or presentation.[docs.perplexity]
Finally, the continuing rollout of agentic “computer‑use” features (like Perplexity Personal Computer and Claude’s computer‑use skills) means you can start offloading more of the mechanical parts of research—clicking through catalogs, downloading images, renaming files, building logs—while you stay focused on interpretation and proof arguments. Think of this week as a nudge to formalize your division of labor: humans handle evidence analysis; agents handle legwork.[youtube][marketingprofs]
NOTE: Claude Fable 5 (in brief)
Anthropic released Claude Fable 5 on June 9, 2026 as its first “Mythos‑class” model that regular users can access broadly through the Claude chat interface, Claude Code, and the Claude API.
It is described as Anthropic’s most capable generally available model, matching the private Mythos 5 model in performance while adding stricter safeguards in high‑risk domains.
Fable 5 supports a 1‑million‑token context window with up to 128,000 output tokens, making it well suited to long‑horizon reasoning and complex, multi‑step tasks such as large research syntheses.
Pricing is currently about double Claude Opus (around 10 USD per million input tokens and 50 USD per million output tokens), but Anthropic is temporarily including Fable 5 at no extra cost on Pro, Max, Team, and certain enterprise plans through June 22 before switching it to usage‑credit billing.
Why it matters for genealogists
With its huge context window and stronger long‑form reasoning, Fable 5 looks especially promising for:
Ingesting a large body of notes and transcriptions (for example, a complete county research file) and helping you draft narrative reports or proof arguments without losing track of nuance.
Orchestrating more complex, agentic workflows, where the model has to plan steps, check its own work, and update a plan over time as you feed it more evidence.
C. Plug‑and‑play AI micro‑workflows to try today
Each of these is tied directly to one of the releases or clarifications above. You can implement them in whatever front‑end you prefer (Perplexity, ChatGPT, Claude, Gemini, etc.) as long as it offers the named model or capability.[familysearch]
1–5: Model‑comparison and method‑testing (Perplexity cookbook + June model guidance)
Test locality‑context quality across models (Perplexity model comparison)
Use the Perplexity multi‑provider tool to send one prompt—“Produce a concise locality guide for probate and land records in Oklahoma Territory, 1890–1907”—to GPT‑5.4, Claude Sonnet 4.6, Gemini 3.1 Flash‑Lite, Grok 4.20, and Perplexity Sonar simultaneously.[docs.perplexity]
Compare which model finds more record types, cites better repositories, and structures the guide in a way you would trust for real work.
Benchmark research‑plan drafting for a real brick wall
Pick one research question (for example, “Identify parents of John Doe, b. 1872 Tennessee, in Indian Territory by 1900”) and send the same prompt to the same set of models via the comparison tool.[docs.perplexity]
Evaluate which model gives a plan closest to BCG‑style standards, then standardize on that model for plan drafting in your practice.
Evaluate summary fidelity of a long case study (Claude Sonnet 4.6 vs GPT‑5.5)
Paste a published proof argument or long research report into a tool that supports both GPT‑5.x and Claude Sonnet 4.6.[marketingprofs]
Ask each model for a 300‑word summary and a bullet list of key negative searches, and see which retains more nuance and caveats.
Choose a default “quick‑answer” model for your daily log
Following June guidance, set GPT‑5.5 Instant (or equivalent) as your default for short, interactive questions inside your research log tool (Notion, Obsidian, Airtable Omni, etc.).[felloai]
Use it all week for quick locality checks, glossary questions, and high‑level timelines, while reserving Claude or Gemini for heavier tasks.
Compare photo‑analysis responses (Gemini vs others)
Upload the same scanned family photo (with inscriptions or signage) to a Gemini 3.1‑based tool and a GPT‑5.x or Claude tool if they support images.[marketingprofs]
Ask each to describe visual clues and hypothesize a date range and location; adopt the model that gives the most grounded, evidence‑based reasoning for ongoing photo work.
6–10: Agentic “computer‑use” tasks (Perplexity Personal Computer, Claude Sonnet 4.6)
Automated download and naming of new record images (Perplexity Personal Computer)
On a Mac with Perplexity Personal Computer, create an instruction: “Whenever I download a census or vital‑record image, rename it to ‘Surname‑Firstname‑RecordType‑Year‑Jurisdiction’ and move it into my Genealogy\Images folder.”[youtube]
Test it while working through a batch of Ancestry or FamilySearch images.
Batch creation of source‑citation stubs from a folder of PDFs (Claude computer use)
With Claude Sonnet 4.6’s file and computer‑use skills, point it at a folder of downloaded record images or PDFs.[marketingprofs]
Ask it to extract basic citation details (collection, repository, image number, date) and draft a table that you can paste into RootsMagic or Zotero.
Browser‑driven catalog skimming for locality research (Perplexity Personal Computer)
Instruct Personal Computer: “Open the FamilySearch catalog for [county], search for probate and land records 1850–1930, and paste all relevant collection titles, date ranges, and film notes into a Google Doc.”[youtube]
Review and clean up the generated locality resource list instead of collecting it manually.
Automated log of research steps in a client project (Claude Sonnet 4.6)
While you work in your browser, occasionally ask Claude (with computer‑use enabled): “Summarize what I have done in the last 30 minutes and draft a bullet‑point research log entry.”[marketingprofs]
Paste that log into your formal research notes or time sheet.
Folder triage for inherited digital files
Point Claude or Perplexity’s computer‑use agent at a messy folder of scanned documents from a cousin.[youtube][marketingprofs]
Ask it to group files into subfolders by record type (census, vital, land, church, military) and produce a manifest listing file name, inferred record type, and date.
11–15: Transcription, translation, and AI‑indexed records (Gemini, platform AIs, FamilySearch AI overview)
First‑pass handwriting transcription of difficult records (Gemini 3.1)
Upload an image of a 19th‑century probate, church register, or land deed into a Gemini‑based tool known for strong multimodal transcription.[familysearch]
Ask for a line‑by‑line transcription in the original language, plus a plain‑language English paraphrase you can annotate and verify.
Quick translation of foreign‑language records (FamilySearch AI + general models)
When working with foreign‑language parish or civil registers, run the image through your preferred AI (GPT‑5.5, Claude, or Gemini) for a draft translation, then compare key names, places, and dates against what FamilySearch’s AI‑indexed record or help text suggests.[familysearch]
Use discrepancies as a checklist of items to verify.
AI‑assisted search refinement using AI‑indexed records
Start with an AI‑indexed collection (for example, a FamilySearch full‑text collection) and copy several index entries for your target surname.[familysearch]
Ask GPT‑5.x or Claude to analyze common spelling variations, locations, and occupations, then feed those variants back into the site’s search filters.
Cross‑checking AI‑suggested tree hints
When Ancestry, FamilySearch, or MyHeritage suggests records or relatives based on AI matching, paste the hint details into GPT‑5.5 or Claude.[felloai]
Ask the model to list potential conflicts (dates, geography, relationships) and suggest what additional evidence you should seek before accepting the hint.
Audio or video oral‑history transcription
Upload an MP3/MP4 of a family interview to a Gemini or Perplexity‑backed tool that supports native audio transcription, then ask GPT‑5.x or Claude to turn the transcript into a structured research note: people mentioned, places, time periods, and unresolved questions.[familysearch]
16–20: Writing, story‑craft, and teaching (GPT‑5.5, Claude Sonnet 4.6, Grok 4.2)
Immediate research‑question refinement (GPT‑5.5 Instant)
Paste your draft research question for a client or project into GPT‑5.5 Instant.[felloai]
Ask it to refine the question to be specific, time‑bound, and testable, plus suggest 3–5 sub‑questions you should track in your log.
Turning dry notes into narrative reports (Claude Sonnet 4.6)
Paste a bullet‑point research log into Claude Sonnet 4.6.[marketingprofs]
Ask it to generate a polished, source‑aware narrative account (with clear “Findings,” “Analysis,” and “Next steps”), which you then fact‑check against your citations.
Historical‑context sidebars for reports or blog posts (any flagship model)
Use GPT‑5.5 or Claude Sonnet 4.6 to produce short historical‑context sidebars (for example, “What did land acquisition look like in Oklahoma Territory in the 1890s?”) for inclusion in reports or blog posts.[felloai]
Ask explicitly for citations and a focus on record‑creation processes rather than generic history.
Multi‑agent “devil’s advocate” review of a conclusion (Grok 4.2)
Paste a proof summary or key conclusion into Grok 4.2, which uses multiple internal agents to debate and synthesize an answer.[marketingprofs]
Ask it to argue against your conclusion and list alternate hypotheses and missing evidence, then decide whether any of those gaps are real.
Workshop demos comparing model styles (Perplexity model comparison + June guidance)
For a class or society webinar, use the Perplexity comparison tool to show how GPT‑5.4, Claude Sonnet 4.6, Gemini 3.1, and Grok 4.20 answer the same genealogical prompt (for example, “Create a research plan for a man who disappears between the 1900 and 1910 census”).[docs.perplexity]
Use the differences to teach participants how to evaluate AI output critically instead of trusting any one model.
No comments:
Post a Comment