The Model Leapfrog
A dated, checked timeline of the 2025-2026 model releases from Anthropic, OpenAI, Google, xAI and others, what each meant for shluchim, and how the group's 'which model for what' advice changed.
On this page
"Which model is best right now?" is the question the group asks most. The honest answer has changed every few months since the group started in July 2025, because the big labs keep passing each other. Members call it the leapfrog. By September 2026 one member summed it up as "one day this AI agent is the one to be at; the next day it's a different one."
This page lists the releases that mattered, with dates checked against each company's own announcements, and what the group made of them at the time.
Where things stand, October 21, 2026
| Lab | Current top models | Notes |
|---|---|---|
| Anthropic (Claude) | Fable 5.1, Opus 5.5, Sonnet 5.5, Haiku 5.5 | All four shipped between Sep 15 and Oct 7, 2026. Opus 5.5 is the everyday heavy hitter; Fable 5.1 is stronger and uses much more of your allowance. |
| OpenAI (ChatGPT) | GPT-6 Astra, GPT-6.1 Sol, GPT-6 Luna, plus the GPT-5.6 family | Free users get GPT-6 text chat. Astra and the Sol models need a paid plan; Dots (the always-on agent) is on Pro only. |
| Google (Gemini) | Gemini 4 Argon, Gemini 3.1 Pro, 3.x Flash models | Gemini 4 Argon reached Google AI Pro and Ultra subscribers on Oct 13, 2026. |
| xAI (Grok) | Grok 4.7 | Released Sep 21, 2026. The company's announcements on x.ai now carry the name SpaceXSI. |
The timeline
Dates are the release or announcement date from the company's own page. The right-hand column is what the group said, with the date of the chat.
| Date | Lab | Release | What it meant for shluchim |
|---|---|---|---|
| Aug 2025 | OpenAI | GPT-5 | The group watched its first big launch live. By early September members complained it felt slower; one noted "they are always tweaking models under the hood." |
| Nov 2025 | Gemini 3 Pro | The group admin moved from ChatGPT to Gemini. Members were impressed by its Chassidus answers and visual layouts. | |
| Nov 20, 2025 | Nano Banana Pro (Gemini 3 Pro Image) | Became the flyer-background favorite, with better text inside images. | |
| Nov 24, 2025 | Anthropic | Claude Opus 4.5 | In December a member reported Opus 4.5 fixing a raffle app another model had broken. |
| Dec 11, 2025 | OpenAI | GPT-5.2 | In January the admin called it "really strong" and said ChatGPT's new image model beat Nano Banana Pro for ads and edits. |
| Jan 16, 2026 | OpenAI | ChatGPT Go at $8 in the US | A cheaper tier between Free and Plus. It may show ads. |
| Feb 5, 2026 | Anthropic and OpenAI | Claude Opus 4.6 and GPT-5.3-Codex, same day | The admin said to brace for more launches that month. A long debate followed on whether shluchim should bother learning AI at all. |
| Feb 17, 2026 | Anthropic | Claude Sonnet 4.6 | Became the default for Free and Pro users. |
| Feb 19, 2026 | Gemini 3.1 Pro | Since January members had been told to pick "Pro" in the Gemini menu, because the Fast and Thinking options ran Flash models that made more mistakes. | |
| Feb 2026 | Nano Banana 2 (Gemini 3.1 Flash Image) | A faster image model in the Gemini app. | |
| Mar 5, 2026 | OpenAI | GPT-5.4 | Little discussion in the group. |
| Mar 24, 2026 | OpenAI | Sora app shutdown announced | Video makers moved to other tools. See Sora. |
| Apr 16, 2026 | Anthropic | Claude Opus 4.7 | Early reports: strong first answers, weaker late in long chats, more tokens used. |
| Apr 17, 2026 | Anthropic | Claude Design | A design tool for slides, one-pagers and visuals. Some members dropped Gamma for it by June. |
| Apr 23, 2026 | OpenAI | GPT-5.5 | In May a member called GPT-5.5 inside Codex the best coding setup; others still preferred Claude. |
| May 6, 2026 | Anthropic | Higher limits after a compute deal with SpaceX | Claude Code's five-hour limits doubled on paid plans. The admin said low limits had been "the #1 reason not to use Claude." |
| May 19, 2026 | Gemini 3.5 Flash, Gemini Omni (video), Gemini Spark (agent), a $100 AI Ultra tier | Spark started as a US beta for Ultra subscribers. | |
| May 28, 2026 | Anthropic | Claude Opus 4.8 | Became the model that flagged requests fall back to. |
| Jun 9, 2026 | Anthropic | Claude Fable 5 (and Mythos 5 for restricted partners) | Included in paid plans at no extra cost through June 22, then on usage credits. A member's comment: most tasks will not need a model this advanced. |
| Jun 12 to Jul 1, 2026 | Anthropic | Fable 5 suspended for everyone under US export controls, then redeployed | Members who had built workflows on Fable fell back to Opus for three weeks. |
| Jun 23, 2026 | Anthropic | Claude Tag (Claude in Slack) | Builders compared it with self-hosted agents. See Claude Cowork and agent tools. |
| Jun 30, 2026 | Anthropic | Claude Sonnet 5 | Became the Free and Pro default. The admin: more quality inside the same limits. |
| Jul 9, 2026 | OpenAI | GPT-5.6 in three tiers (Sol, Terra, Luna) and ChatGPT Work | Agents that members switched to a new model name broke until the tools caught up. In August members passed around a tip that Luna at maximum effort is nearly as good as Sol for much less. |
| Jul 2026 | xAI | Grok 4.5 | A member said it was "very good and very cheap"; another: the first time in a while anyone besides Anthropic or OpenAI was in the conversation. |
| Jul 21, 2026 | Gemini 3.6 Flash | ||
| Jul 24, 2026 | Anthropic | Claude Opus 5 | Anthropic said it comes close to Fable 5 at about half the price. One member: "it beats Fable." Several members moved all their agents to it. |
| Aug 12, 2026 | xAI | Grok 4.6 (and Grok Bot, a team of always-on agents, Aug 11) | Grok Bot was named alongside other personal agents but not widely tried. |
| Aug 13, 2026 | Gemini 3.7 Flash | Now powers Gemini Spark for Pro and Ultra subscribers. | |
| Sep 3, 2026 | OpenAI | GPT-6 Astra | The admin, in September: "Fable 5.1 and GPT 6 Astra. A great week for AI." |
| Sep 8, 2026 | Meta | Muse, a personal AI agent | US launch, free core service, works inside WhatsApp. See Personal AI Agents. |
| Sep 15, 2026 | Anthropic | Claude Fable 5.1 (and Mythos 5.1) | Members took it up within days. Quality praise, but "Fable 5.1 is draining me" and "maxed out two accounts" were common. |
| Sep 21, 2026 | xAI | Grok 4.7 | |
| Sep 22, 2026 | Anthropic | Claude Opus 5.5 | Anthropic said it matches Fable 5.1 on most work at lower cost. Members used it to edit an event video from 73 clips and to build websites from reference images. |
| Sep 28, 2026 | Anthropic | Claude Sonnet 5.5 | |
| Sep 29, 2026 | OpenAI | GPT-6.1 Sol and Dots (DevDay) | Among individual plans, Dots, OpenAI's always-on agent, is listed for Pro only. |
| Oct 7, 2026 | Anthropic | Claude Haiku 5.5 | |
| Oct 13, 2026 | Gemini 4 Argon | Rolled out to Google AI Pro and Ultra. Too new for a group verdict as of this review. |
Other labs came up less often. A builder ran his agent on Chinese GLM models through z.ai to save money in spring 2026; Qwen keeps a following for Hebrew and Yiddish translation.
The group's "which model for what" advice, as it changed
- August 2025. In a three-model test, Claude was judged the most accurate on technical Chassidus questions, with the caveat that it needed more prompting to sound popular rather than academic.
- September 17, 2025. The admin: "Every shliach must have at least one paid AI plan," starting with ChatGPT Plus. A same-day poll: 24 paid, 13 free only.
- October to November 2025. The admin moved to Gemini and called it the best. A member's half-joking rule: "for 90% of us, 90% of use cases return results 90% similar in quality across the four major models."
- January 9, 2026. "Stick with the big three: Gemini, ChatGPT, and possibly Claude. Pay for one and use it for everything."
- January 21, 2026. Updated advice: ChatGPT or Gemini for phone and day-to-day use; ChatGPT's image model for ads and edits; Gemini if Gmail and Docs matter; Claude for serious desktop work, with the lowest Pro limits of the three. In Gemini, always pick Pro.
- March to April 2026. The admin: "Claude is the best model (in my and many others' experience), its biggest chisaron is it has much tighter usage limits than ChatGPT and Gemini." ChatGPT remained the pick for phones and images.
- May 2026. After the May 6 limit increase, the admin said the $20 Claude plan might now be enough for many.
- June 2026. Claude tier rule: Sonnet for plain writing, Opus for complex reasoning and long documents, Fable only for very long work that must stay consistent.
- July 20, 2026. The admin: ChatGPT "has caught up majorly." ChatGPT Plus for most beginners; Claude if you are coding or running agents. By August 6 it was "preference."
- August to September 2026. The admin told people to "get off Gemini," saying it had fallen behind. Builders set a rule that Claude Code sub-agents never run on Fable, to save usage.
- September 29, 2026. "You don't need to constantly be switching. Stick to one and get really good at it. When its deficiencies really bother you, switch." Several members agreed.
- October 2026. Opus 5.5 became the default recommendation for site building and long Claude work. Gemini 4 Argon had just landed and had not been tested by the group.
What has held up
The labs copy each other's features within weeks. Chat, then file tools, then agents that work on your computer (Claude Cowork, ChatGPT Work, Gemini Spark) all arrived at roughly the same time in 2026. Picking a lab matters less than learning one well.
Model names change faster than the skills do. Clear instructions, giving the model your sources, and checking its work carry over to whatever ships next.
An older group poll on what matters most in a model ranked "remembers and understands your context" first (50 votes) and "being able to trust it" second (39). Coding and agent ability ranked near the bottom (6). Most members use AI for writing, research and admin, not engineering.
How to size up a new model
A routine a member shared in July 2026:
Here is the official announcement for [model name]. Read it and tell me,
in plain words: what is this model best at, what is it worse at than the
model I use now ([current model]), and does it cost more of my usage?
Then write me a short prompt I can use to test it on my own work:
[describe a real task you do every week].
Another member asks the assistant to look at his own chat history and recommend which model and effort level fit the work he actually does. When a new model arrives, existing setups can break, so test it on a copy of an important workflow before switching everything over.
Related
- Claude, ChatGPT, Gemini, Grok
- Paid vs. Free, where model choice meets the price tier
- AI and Torah Accuracy, where the model matters less than the sources you give it
- Personal AI Agents
- Toolbox