Here’s your compact, “what actually changed this week and what to do with it” briefing for working genealogists and family historians based on releases and trackers updated through July 17–18, 2026.
A. Named releases & features (last ~72 hours + still hot)
Moonshot AI – Kimi K3 (frontier, open-weight, long‑context)
New reasoning‑focused model released July 16, 2026 with strong performance on long documents and complex analysis, positioned as a frontier‑level open‑weight option.[aireleasetracker]OpenAI – GPT‑5.6 family (Sol, Terra, Luna) GA in APIs and ChatGPT
The full GPT‑5.6 line moved from staged rollout to broad availability in early–mid July, with Luna as the smallest, cheaper model and Sol/Terra as larger reasoning models.[llm-stats]OpenAI – GPT‑5.6 Luna pricing + long‑context adjustments
Luna is priced around $1 per million input tokens and $6 per million output, part of an updated pricing structure that makes extended context (128k+) more affordable.[kersai]OpenAI – GPT‑5.6 Sol/Terra (research‑grade reasoning)
Sol and Terra are higher‑capacity models aimed at deep research, multi‑document synthesis, and more reliable complex reasoning, now fully available in production.[llm-stats]OpenAI – GPT‑Live full‑duplex voice rollout (still expanding)
GPT‑Live, OpenAI’s real‑time, full‑duplex voice assistant built on GPT‑5.x, continues rolling out; it allows conversational exploration of research, timelines, and sources via voice.[skycrumbs]OpenAI – GPT‑5 Mini updates (better structure/output)
The lightweight GPT‑5 Mini tier got upgraded weights in early July with improved instruction‑following and structured output, useful for fast, cheap list‑building and checklists.[skycrumbs]OpenAI – Open‑weight models gpt‑oss‑120b and gpt‑oss‑20b (recent)
OpenAI has introduced two open‑weight models (gpt‑oss‑120b and gpt‑oss‑20b), giving researchers and tool builders high‑quality models whose weights can be hosted and customized.[help.openai]Anthropic – Claude Sonnet 5 (new mainline model, late June but “current”)
Claude Sonnet 5 landed at the end of June and is now a primary, widely‑used model with strong narrative, summarization, and long‑context capabilities.[llm-stats]xAI – Grok 4.5 (reasoning + lower prices)
Grok 4.5 launched around July 8, 2026 and is priced at roughly $2 per million input tokens and $6 per million output, with a focus on reasoning and high‑throughput workloads.[llm-stats]Meta – Muse Spark 1.1 (multimodal, model API)
Muse Spark 1.1, released July 9, 2026, exposes a public API for a multimodal model good at images and text, adding another strong option for integrated document + image workflows.[aireleasetracker]Google – Gemini 1.5 Pro (1M‑token context, multimodal)
Gemini 1.5 Pro, already in wide release by mid‑2026, offers a 1‑million‑token context window, fast generation, and strong multimodal support; it remains an important long‑context tool this week.[howdoiuseai]Open-weight – Kimi K2/K3 line (frontier‑level open models)
Moonshot’s Kimi series (K2 previously, K3 now) represent open‑weight models that match or exceed proprietary models in reasoning and long‑context tasks, giving self‑hosters serious options.[llmgateway]Open-weight – GLM 5.x (coding/agentic support, strong open baseline)
GLM 5.x (and specifically 5.2) is positioned as a leading open‑weight model with strong coding and agent support, often used as the engine behind custom AI genealogy helpers.[openrouter]Perplexity – Research‑focused AI search (ongoing, but critical for this week)
Perplexity continues to operate as a real‑time, citation‑backed AI search engine with a freemium model, widely used for up‑to‑date research and “where are the records now?” questions.[howdoiuseai]ChatGPT platform – Deep Research, Record Mode, memory & connectors
The current ChatGPT platform bundles deep multi‑source research, meeting transcription/summary (Record Mode), persistent memory, and 60+ app connectors—including Google Drive and Slack—into one environment.[howdoiuseai]Gemini – Thinking/transcription modes for long, handwritten documents
Gemini’s current generation (often referenced as “Gemini 3.0” in genealogy usage guides) adds improved thinking modes and handwriting‑friendly transcription, particularly useful for deeds and wills.[denyseallen.substack]Claude – Tool‑using assistant with long‑document timelines
Claude’s tool‑use features and strong timeline generation capabilities make it a natural choice for converting long narrative genealogical documents into clean chronological timelines.[youtube][marketingaiinstitute]Family history AI ecosystem – FamilySearch AI Research Assistant (ongoing)
FamilySearch’s AI Research Assistant uses modern LLMs to provide guided record‑finding, summarization, and translation within FamilySearch’s ecosystem, and continues to be updated.[youtube][familysearch]Pricing landscape – Cheaper long‑context across frontier and open‑weight
Across GPT‑5.6, Grok 4.5, Muse Spark 1.1, Kimi K3, and others, prices per million tokens for long‑context reasoning have dropped, making multi‑document uploads and long projects more economically viable.[kersai]
B. Implications for genealogists this week
For working genealogists, the big story is that long‑context reasoning just got cheaper and more widely available across both proprietary and open‑weight models. That means you can realistically feed an entire locality study, multi‑generation timeline, or a full probate packet into a single session and ask for structured analysis, instead of chopping everything into tiny pieces.[rauljitechnologies]
At the same time, models are differentiating by role: ChatGPT (GPT‑5.6) for planning and structured workflows, Perplexity for citation‑backed “where are the records now?”, Gemini for transcription and multimodal document work, Claude for timelines and narrative polish, and Grok/Kimi/GLM for those who want open‑weight or lower‑cost custom agents. You benefit by matching the task to the model instead of trying to force one tool to do everything.[openrouter]
Finally, the new GPT‑Live and improved smaller models (GPT‑5 Mini, Luna) make lightweight, conversational, “always‑on” assistance more practical, especially for routine tasks like checklist generation, quick record‑type reviews, and voice‑driven brainstorming while you’re looking at microfilm or digitized images. Combined with FamilySearch’s AI assistant and other domain‑specific tools, you now have an ecosystem where you can move seamlessly from “What records should exist?” to “Where are they?” to “What do they say?” to “How do I write this up?” inside coordinated AI workflows.[familysearch][youtube]
C. Plug‑and‑play AI micro‑workflows you can try today
Below are at least twenty concrete, current, genealogy‑specific workflows tied explicitly to the named releases/features above. I’ll assume you’re comfortable moving between tools; each item is one self‑contained “micro‑workflow” you can run as a small experiment.[familysearch][youtube]
1–5: Long‑context case files with GPT‑5.6 and Gemini
Probate packet deep read (GPT‑5.6 Sol/Terra)
Upload a multi‑document probate file (petition, inventory, accounts, distributions) to ChatGPT powered by GPT‑5.6 Sol or Terra and ask for:A chronological timeline of events
List of all named individuals with roles (executor, heir, creditor)
A summary of property categories (land, livestock, household goods)
Use the cheaper long‑context pricing to keep the entire packet in one run.[huggingface]
Locality guide synthesis (GPT‑5.6 Luna + Perplexity)
First ask GPT‑5.6 Luna in ChatGPT for a “record‑type checklist” for, say, Oklahoma Territory circa 1890–1907 (land, probate, court, tribal, etc.). Then paste each record type into Perplexity and ask, “Where can I access [record type] for [county/time] today?” to get up‑to‑date repositories and URLs with citations.[skycrumbs]Multi‑family cluster reconstruction (Gemini 1.5 Pro)a
Load a large batch of census pages and city directory scans into Gemini 1.5 Pro’s long‑context environment, and ask it to:Identify all households with the target surname
Flag recurring neighbors and addresses
Suggest possible cluster families based on proximity and shared occupations.[aireleasetracker]
Handwritten deed transcription (Gemini “thinking” mode)
For a hard‑to‑read handwritten land deed, upload the image to Gemini and use its advanced transcription/thinking mode to produce a text version. Ask follow‑ups: “Extract grantor, grantee, legal description, consideration, witnesses, and dates in a simple table.”[youtube][familysearch]Narrative source comparison (GPT‑5.6 Sol + Claude Sonnet 5)
Use GPT‑5.6 Sol to produce a structured comparison of several conflicting birth records (church register, delayed certificate, family Bible) in a table (source, date, informant, reliability notes). Then send that table to Claude Sonnet 5 and ask it to draft a narrative “evidence summary” paragraph using genealogical proof‑style language.[marketingaiinstitute][youtube]
6–10: Timelines, conflict tables, and research planning
Automated ancestor timeline (Claude Sonnet 5)
Paste a long, existing narrative biography plus extracts from key records into Claude Sonnet 5 and ask: “Convert this into a chronological timeline of dated events, with columns for event, date, place, source, and notes.”[youtube][rauljitechnologies]Research conflict matrix (Claude + GPT‑5 Mini)
In ChatGPT using GPT‑5 Mini, quickly list all conflicting claims about an ancestor (birth years, parents, residences) and format them as a rough table. Paste that table into Claude Sonnet 5 and ask it to expand into a full conflict matrix with reliability assessment and “next steps” research questions.[marketingaiinstitute][youtube]County‑level research plan (ChatGPT GPT‑5.6 Luna)
Ask ChatGPT with GPT‑5.6 Luna: “Act as a genealogy research planner. Create a 4‑week research plan for tracing land and probate records for [county] in [time period], including archives, catalog searches, and likely record series.” Use the improved structured‑output to get tasks grouped by week.[skycrumbs]Multi‑generation migration path (GPT‑5.6 Terra + Gemini maps/images)
Provide GPT‑5.6 Terra with a list of locations and dates for a family across several censuses and land records. Ask it to infer migration routes and possible transportation corridors. Then use Gemini to generate a simple, labeled map illustration you can include in your report or blog.[familyhistorystorytelling.wordpress]Cluster research brainstorming (Grok 4.5)
Feed Grok 4.5 a description of a “brick‑wall” ancestor and the cluster around them (neighbors, associates, witnesses). Ask it to propose at least ten cluster‑based research avenues (church membership lists, school records, occupational guild files, etc.), ranking them by probability and effort.[familysearch]
11–15: Record‑finding and repository work
Where‑are‑the‑records map (Perplexity + GPT‑5.6)
Take your location and time period, ask GPT‑5.6 Luna in ChatGPT for a list of possible record types (civil registration, court minutes, tax rolls). Then, for each record type, ask Perplexity “Where are [record type] for [county/state, years] currently accessible?” and save the cited URLs into Zotero.[denyseallen.substack]FamilySearch AI Research Assistant scouting
Use FamilySearch’s AI Research Assistant to ask, “Show me potential records for [ancestor] in [region], and summarize why each is relevant,” then immediately test those suggestions against your own hypothesis. Treat it as a scout, not an oracle.[youtube][familysearch]Catalog mining via ChatGPT connectors
With ChatGPT’s connectors to Google Drive or similar, upload an exported catalog search result (CSV or PDF) from a state archive and ask GPT‑5.6 Luna to:Group entries by record series (deeds, judgments, probate)
Flag series that directly mention your surnames
Suggest a visit plan (which boxes to pull first).[howdoiuseai]
Open‑weight “agent” that knows your locality (GLM 5.x or Kimi K3)
If you run your own stack, fine‑tune GLM 5.x or Kimi K3 on a set of locality guides, catalog entries, and your own notes for a single county. Use that agent to answer “Given this research question, which five specific record series in this county should I prioritize?”[openrouter]Repository contact drafts (Claude Sonnet 5)
Paste a description of the records you need and repository details into Claude Sonnet 5, and ask it to draft concise, polite email templates to archivists or county clerks, including call numbers or series names.[marketingaiinstitute][youtube]
16–20: Transcription, photos, and narrative
Batch transcription triage (Gemini + Claude)
Before investing in full transcriptions, upload a bundle of images (probate, land, church registers) to Gemini and ask it only for “keyword extraction” (surnames, place names, occupations). Use those keywords to decide which images go to the front of your queue, then send the most promising items to Claude for full narrative summaries.[youtube][marketingaiinstitute]Photo set clustering and caption help (Muse Spark 1.1)
Use Muse Spark 1.1 to analyze a group of digitized family photos, asking it to cluster images by setting (farm, town, school) and approximate era from clothing/vehicles. Then prompt for suggested neutral captions (“Two unidentified children in rural setting, circa 1910s”) to store in your photo database.[familysearch]Life‑story draft from scattered notes (Claude Sonnet 5 + GPT‑5.6)
Combine scattered research notes into one long document and feed it to Claude Sonnet 5 for a first‑pass narrative biography. Then ask GPT‑5.6 Sol to convert that biography into a research‑style “proof summary” organized by identity, parentage, and migration, citing each claim back to a source list you provide.[rauljitechnologies][youtube]Voice‑driven research brainstorming (GPT‑Live)
Use GPT‑Live via voice while looking at digitized microfilm or working in a reading room. Talk through your current brick wall and ask it to suggest concrete next steps (“three record types in this county I haven’t checked”), capturing the audio or notes as a Saturday‑morning briefing.[huggingface]Automated research log clean‑up (GPT‑5 Mini / Luna)
Export your research log from Zotero, Excel, or RootsMagic as CSV. Ask GPT‑5 Mini or Luna to normalize repository names, flag duplicate searches, and generate a short “this week’s priorities” list based on unresolved items and high‑value record series.[ancestorsandai.buzzsprout]
21–24: Advanced/open‑weight experiments
High‑volume surname sweep (Grok 4.5 or Kimi K3)
If you have a large, text‑based dataset (OCR’d newspapers, tax rolls, court minutes), send it in chunks to Grok 4.5 or Kimi K3 and ask for a structured extraction of all instances of a surname with date/place; then merge into a master spreadsheet.[kersai]Custom “probate explainer” bot (GLM 5.x open‑weight)
Fine‑tune GLM 5.x on a curated set of probate case studies and glossaries. Use it locally as a bot that explains obscure probate terms or procedures in plain language while you’re abstracting wills or estate files.[openrouter]Language‑heavy parish record analysis (Gemini + Claude)
For non‑English parish registers, run first‑pass transcription and translation via Gemini. Then give Claude Sonnet 5 the translated text and ask it to identify naming patterns, godparent clusters, and possible extended kin groups.[familysearch][youtube]“Evidence table” generator across tools (Perplexity + GPT‑5.6)
Use Perplexity to gather up‑to‑date secondary sources on a local event (e.g., land run, epidemic, migration wave) that affected your ancestor’s locality. Then feed those citations plus your primary sources into GPT‑5.6 Terra and ask it to produce a table showing how each source supports or contradicts your current hypothesis.[aigenealogyinsights]
No comments:
Post a Comment