Conol 1.2.0 — Know where your credits go

Credits are how Conol charges for the agent's work, and until now the only place you could see them was a single balance in Settings. This release makes spend visible where it happens: a per-chat history on the Billing page and a running counter in the chat itself. It also adds four models, including Anthropic's new low-cost tier and Google's new flagship, brings voice input back after our speech provider disappeared, and corrects a billing rate on Claude Sonnet 5.5.

Spending history on the Billing page

Why. One number for your balance tells you how much is left, not where it went. People asked the same two questions: which chat used all those credits, and was that long research task worth what it cost? Sub-agents made it worse, because a chat that delegates work to helpers spends credits under several sessions, and none of them showed up as one line anywhere.

How to use it. Open Settings → Billing and scroll to Spending history. Every chat that used credits in the period is listed, most recently charged first, with the total credits, the number of charges, when it was last charged, and a link that opens the conversation. Sub-agents count toward the chat that started them, so a delegated task shows as one row with its full cost. The period switcher offers the last 7, 30 or 90 days or all time, and the total for the period sits above the list. Long histories page with Load more. Charges whose chat was deleted, or that predate per-chat tracking, are grouped under Earlier or deleted sessions so the total still adds up.

Credits used, right in the chat

Why. The history answers "what did this cost" after the fact. While a long task is running, you want the same answer live, without leaving the conversation.

How to use it. The chip in the composer header shows how many credits the open chat has used so far, sub-agents included. It refreshes after each turn settles, with a short lag for the billing queue, and shows one decimal while the number is small so a cheap chat never reads as zero. Hover it for the explanation. Charges from before this release are attributed through the chat's own session, so a helper's work from before the release may be missing from an older chat's count.

Claude Haiku 5.5, Gemini 4 Argon, Gemini 3.8 Flash and DeepSeek V4.1 Flash

Why. Two of the models in the picker had quietly gone stale: Google had retired Gemini 3.5 Flash upstream and was serving its successor in its place, and DeepSeek had retired the Flash models we offered. Meanwhile Anthropic shipped a Haiku that is ten times cheaper than the last one and stronger on agent tasks, and Google shipped a new flagship. We audited the whole catalog against the vendors' own pricing pages and live endpoints, and this release is the result.

How to use it. Open the model menu in the composer. The Popular tab is now a shortlist of today's headliners, one or two per vendor; everything else is still under its vendor's tab.

• Claude Haiku 5.5 (Anthropic): the fast, low-cost tier, at a twentieth of Sonnet 5.5's price. It also now writes your chat titles and the summaries behind the agent's web reading, which used to run on Haiku 4.5. Note that Anthropic charges five times more for a request whose prompt exceeds 100K tokens; Conol compacts the conversation before that point, and bills the higher rate on the rare turn that still crosses it.
• Gemini 4 Argon (Google): the new flagship, strong on long documents and multi-step work. Priced like Opus 5.5 on paid plans.
• Gemini 3.8 Flash (Google): replaces Gemini 3.5 Flash. Chats and saved defaults that pointed at 3.5 Flash move over automatically.
• DeepSeek V4.1 Flash: the current DeepSeek Flash, now with image input. It replaces DeepSeek V4 Flash and DeepSeek V4 Flash Vision, which DeepSeek has retired; existing chats move over.
• DeepSeek V4 Pro and MiMo V2.6: the same entries, now served by their current versions. DeepSeek V4 Pro's price moves to DeepSeek's list price; MiMo V2.6 Pro and Flash keep their prices and gain image input.
• Kimi K3: priced at Moonshot's current list price, slightly higher than before.

Fixes

• Voice input works again. Dictating into the composer, and voice messages sent through Telegram and WhatsApp, stopped being transcribed when the speech model we used lost its last serving host. Transcription now runs on ElevenLabs Scribe, which handled our English and Mandarin tests word for word, with Whisper standing in automatically if Scribe is ever unavailable. Nothing to configure.
• Cached input on Claude Sonnet 5.5 was billed at twice its rate. Since Sonnet 5.5 arrived on September 30, the part of each turn served from the prompt cache was charged at Sonnet 5's cache rate rather than Sonnet 5.5's, which is half as much. Cache reads make up most of a long chat's input, so this mattered more than it sounds. The rate is corrected from this release.