Release Notes
Last updated: October 29, 2026
What's new and what's been fixed, in the order it shipped. This covers user-facing changes only. Not every internal fix makes the list.
October 29, 2026
- ImprovedWhen you say yes to a character checking in, the email now comes the next day instead of a few days later.
- FixedA new account's first three days of unlimited talks now really start: until now they were not applied, so a second talk on the first day hit the daily limit.
October 21, 2026
- ImprovedSay yes once to a character checking in by email, and it will check in a few days after each later talk too. You can skip one in the chat, and every email has a link that stops them.
- ImprovedThe characters keep to your language even when they answer each other, and a talk started from the home page goes on right away with a question to you.
October 16, 2026
- NewNew accounts can talk as often as they like for their first three days, the same as Parley Plus. Near the end you get one note in the chat and one email; after that it is one talk a day again, and nothing you said is lost.
- NewFounding members: the first 20 people who post about Parley on social media (with #ad) and paste the post's link on the Plus page start Plus at 80% off for six months ($1.00 a month, then $4.99). The page shows how many places are left.
- ImprovedLinks to Parley now preview in the language of the page, and the home-screen icon fits round icon shapes on Android.
October 15, 2026
- NewParley Plus: talk over as many things a day as you like, and go back to an earlier talk, for $4.99 a month (¥480 in Japan, ₩4,900 in Korea). The same four answer you. Get it from the note that appears when today's talk is used up, or from Settings → Plan.
- ImprovedAnswers in a talk now come in one at a time and are written out as you watch, the way a group chat reads, instead of four landing at once.
- ImprovedThe characters keep to the language you write in. What Parley remembers about you now carries the day you said it, so an old plan is not taken for today's, and a talk you mark as a test is not remembered.
October 12, 2026
- NewOn the free plan you can now talk one thing over every day: your question and five follow-ups. It no longer draws on the one-time trial Credit, which stays for Rooms and Research. A new talk opens at 00:00 UTC.
- NewToday's dilemma (/today): one small decision a day that anyone can answer, no account needed. Pick a side first, then see what the four characters said and how everyone else voted. The dilemmas are written by Parley and never taken from anyone's conversation. Only the day's two totals are stored.
- NewConsult now has a cast. Four characters answer you, each run by a different company's AI and each with a way of looking at things: Spark says try it, Pause says wait, Cozy asks how you feel, Frank says it straight. Their faces change with what they say. Model names no longer appear in the chat.
- NewTidy sums up where the talk stands: what the four agree on, where they split, and what it hinges on. It speaks up every few messages, or whenever you ask.
- ImprovedThe app now opens on Consult. Checking an AI answer, Research, Projects and the rest are one click away under "More" in the sidebar.
- ImprovedAnswers come in the language you wrote in, and Parley brings up something it remembers about you only when it bears on the question.
- ImprovedNew icon and link preview, and a home page that starts with today's dilemma.
October 9–10, 2026
- NewConsult (/app/consult): write what is on your mind, and several AIs answer like friends in a group chat, each with where it stands, then push back on each other. After the first answers, two reply to each message and one more may chime in. Parley remembers the basics of your situation on its own, so you do not have to explain it again; you can see, delete, or turn it off under "What we remember". Health, sex and money details are kept only when you say "remember this", and ID numbers and passwords never are.
- ImprovedPersonal questions (relationships, career, family) get a clear answer instead of a critique: each model says which way it leans, why, and what would change its mind, so you can see where they really differ.
- ImprovedGemini 3.7 Flash, retired by Google, is no longer in the model list. Rooms that already use it keep working: Google now answers them with Gemini 3.8 Flash at the same price.
- NewParley works inside Claude Code, Claude Desktop and Cursor. Create a key in Settings → Claude Code / MCP, and the AI you work with can ask models from three companies for a second opinion and get back where they agree and split. Setup takes one line; see the guide at /mcp.
- ImprovedPanel answers now open with the weakest point instead of praise. Models were agreeing with each other far more often than the questions deserved; now each one names what is wrong first, and says what it checked if nothing is.
- ImprovedOn the free trial, choosing a model the remaining Credit cannot cover no longer stops the reply. The same company's low-cost model answers instead, and the Room says so once.
- ImprovedPanels pick companies in a fixed order (Anthropic, OpenAI, Google first) and no longer use Command R7B, which tended to summarize the question rather than answer it.
- ImprovedLarge images are scaled down before upload, so they send faster and cost less to read.
- FixedNew accounts start in the language of the page they signed up from, and Chinese text is no longer mistaken for Japanese.
October 5, 2026
- ImprovedAsking something now starts with the question. "Ask a question" opens a box you type into, and the Room is created when you send it. Choosing which models answer is still there, one click down, and no longer stands between you and asking.
- ImprovedA new account arrives to a Room that is already being answered. Three models from three different companies take the same question independently, so the thing Parley does is visible before you have set anything up.
- ImprovedWhen a model cannot answer, the message now says what happened and what to do about it, in the language of the Room. On the free and Credit plans it no longer talks about API keys, which are ours rather than yours.
- FixedA failure that could never have succeeded — a rejected key, a spent quota, a retired model — is now reported straight away. It used to be retried several times first, which looked like nothing happening for minutes.
- FixedThe free trial Credit goes roughly three times further. Each reply was reserving far more budget than replies actually use, so a question asked of three models could exhaust the whole trial at once.
- ImprovedGames are no longer offered when creating a Room. Existing game Rooms still work.
September 29, 2026
- FixedSolar is temporarily not included in the free and Credit plans. Upstage began rejecting every request from our servers, so no Solar reply could get through. It still works with your own Upstage key, and we will restore it to the included set once Upstage lifts the block.
September 24, 2026
- NewNewer models from four providers: GPT-6 Luna, Sol and Astra, Claude Opus 5.5 and Fable 5.1, Gemini 3.8 Flash, and Grok 4.7. Every one was checked against the provider's live API before being listed. GPT-6 Luna and Claude Opus 5.5 are both cheaper than the models they follow.
- NewA comparison now shows where the answers split. Each contested point is laid out with the models grouped by the position they took, so a three-way disagreement reads as three positions rather than as two lists you have to reconcile yourself. It appears on shared pages too.
- ImprovedShared comparison pages read as a report rather than a page of output, and each one gets its own social card.
- ImprovedGPT-5.6 Sol was listed at a higher price than OpenAI actually charges, which made it reserve more Credit than it needed. Corrected.
- FixedA new account is now set up for Credit the moment it first needs it. Previously that only happened on one particular path, so it was possible to be told you had reached a limit before using anything.
- FixedUsage no longer reports 0% used for an account that has no Credit at all. It says so, and says why, instead of showing an empty bar that looks like room to spare.
- FixedThe landing page loads faster. Bot protection used to load on every page including the marketing site, where it did nothing but cost a second of processing.
September 21, 2026
- ImprovedThe Chrome extension's composer now keeps its controls inside the input box — which models to ask, and Summarize this page, sit on the same row as Send instead of in a strip underneath it. In a side panel that strip was a whole row of chrome around a small text area.
- ImprovedA Panel now pages by question. The arrows above the cards move every model's answer and the comparison below them to the same question at once, so a comparison is always the comparison of the answers you are looking at.
- FixedA long answer opened from a Panel card no longer runs past the bottom of the window. It scrolls inside the dialog, on a laptop and on a phone.
- FixedSigning up through the Chrome extension no longer asks for your own API keys. A new account starts on Free with its models already set up, the same as signing up on the web.
September 17, 2026
- NewThe Parley Chrome extension is live on the Chrome Web Store. It opens Parley in a side panel so you can ask several models about the page you are already reading, and it signs in with the same account as the web app.
- NewPanel comparisons can now be shared as a public link. It publishes a read-only page anyone with the link can open without signing in, frozen at the moment you create it, and you can take it down at any time.
- ImprovedCompare answers is available on every plan, Free included. So are the Consensus disagreement breakdown, Decision Record, note summaries and Auto Mode. What a paid plan buys is budget, not access to features.
- ImprovedRoom invites work on every plan. Free accounts can send five in total.
- FixedInvite emails no longer render a Room name or display name as anything clickable.
- FixedA comparison could list a model under a garbled name. Names are now checked against the Room's actual Participants.
September 15, 2026
- ImprovedModelSight is now called Panel. Same Room, plainer name: one question goes to several models, and you read their independent answers side by side.
- ImprovedNew Room now leads with the two things that need no setup, Chat and Panel, with Brainstorming, Research and Games below them. The app home's main button opens a Chat directly.
- ImprovedNew Rooms start with the least expensive models selected, which makes the free trial Credit last roughly ten times longer.
- ImprovedModel lists no longer quote per-token provider prices on the plans that run on Parley's own credits, since those plans are never billed that way.
September 14, 2026
- FixedFree accounts couldn't create a Room with three AI Participants, even though the plan allows three.
- ImprovedThe free plan's Room limit is lifted from 4 to 20, and models on the free and Credit plans are now presented as models rather than as API keys to register.
- FixedFree accounts couldn't add an API key of their own, because the included models were counted against the key limit.
September 12–13, 2026
- NewShared documents are now written section by section, so two models can work on the same document without overwriting each other, and you can ask a second model to review any single section.
- NewExport a document to Word or Google Docs.
- NewResearch portfolio: every Research Room you own in one place, with an evidence timeline showing how an answer changed over time.
- NewOthello, alongside Shogi and Guess My Card, and a stamp collection you fill in by beating each model.
September 6–11, 2026
- NewPeer review and fact-checking: a second model checks a claim before it settles into the conversation.
- NewProjects group related Rooms and share one set of background files across them.
- ImprovedConsensus and Decision are one step now. A decision record is created with the first Consensus and stays editable in place, instead of waiting for a separate button nobody pressed.
- ImprovedThe free plan now allows 3 AI Participants per Room, up from 2.
September 5, 2026
- NewAttach a CSV or Excel file to any conversation so the models can analyze the data directly.
- ImprovedResearch Dashboard now includes an activity log and a sortable evidence table, alongside a clearer up-to-the-minute summary.
- ImprovedPaid-tier Brainstorming and Research conversations now default to stronger models.
- FixedFile uploads (images and data files) occasionally failed due to a permissions issue.
- FixedResearch Rooms could occasionally produce a written document instead of updating the Dashboard.
September 1–4, 2026
- NewCredit-based pricing: Starter, Standard, and Pro plans, each with its own monthly usage budget instead of a feature wall. Every paid tier unlocks Auto Mode, Research, and Brainstorming.
- NewA BYOK plan for bringing your own API key, with no usage cap.
August 31, 2026
- NewBrainstorming and Research can now hand their output to each other, turning a Research report into a Brainstorming starting point or the reverse.
August 29, 2026
- ImprovedResearch Dashboard redesigned with clearer priority ranking and live-update indicators, so it's obvious at a glance what's changed and what matters most.
- NewBrainstorming and Research now suggest a default role for each model based on its real strengths, instead of price alone.
- FixedSmoother, less choppy text rendering across Chat, Brainstorming, and Research.
August 27–28, 2026
- NewKorean language support, including KRW pricing, Kakao Pay, and Naver Pay.
- NewPaste or upload an image directly into any conversation.
- NewEmail notifications for account and billing changes.
- NewAn interactive live demo on the website, so you can try Parley before you sign up.
August 26, 2026
- ImprovedCompare Answers results now stay attached to the Room, with a link back to any past comparison.
August 22, 2026
- NewPLaMo (Preferred Networks, Japan) added, bringing the model lineup to 9 AI providers.
- NewA Best Practices guide for getting the most out of multi-model conversations.
- NewAnnual billing, with two months free.
August 21, 2026
- NewCompare Answers broadcasts one question to every model at once, showing exactly where they agree and disagree.
- NewPro plan launched.
- ImprovedA provider that hits its usage limit or has a rejected key now pauses automatically instead of failing every request.
August 19, 2026
- NewResearch Rooms cluster evidence by claim and score it by independent corroboration across models.
- ImprovedLive web search extended to GPT and Grok, alongside Claude and Gemini.
August 18, 2026
- NewBrainstorming Rooms run a role-based debate that moves through Goal, Divergence, Convergence, and Decision.
- NewAn interface language setting, independent of whatever language a Room's own conversation happens to be in.
August 17, 2026
- NewGuess My Card, the first Parley Game, is a bluffing game played against your own AI models.
August 16, 2026
- NewDeep Research mode for fact-checkable questions, grounded in real citations.
August 14–15, 2026
- NewCatch-ball: when you address multiple models at once, the first to answer gets one extra turn to react to the others without you asking again.
- NewIn-conversation @parley notes and commands.
August 13, 2026
- NewAI-to-AI auto-chaining lets models ask each other direct follow-up questions, continuing a discussion without you in the loop on every turn.