October 2026 · The month's AI updates
Following AI news all through October, one thought kept coming back: what was new this month was mostly not the models, it was the bill. At the start of October, a million output tokens dropped to $0.50 with Claude Haiku 5.5; by the end of the month, GPT-6.1 Terra went from $12 to $5. ChatGPT's free plan moved to GPT-6. Google put its best model behind a paywall and said up front that it will double the price on December 1.
There was plenty of news, but very little of it changes the work on your desk. For this roundup I ran the month through a single filter: does a small business or an employee need to change how they use an AI tool this month? The items that passed made it into the five sections below; the ones that didn't went to "Overhyped." All figures come from official pages; prices are in dollars, as shown on the official pages, and haven't been converted to Turkish lira.
What happened: OpenAI opened GPT-6 to everyone in ChatGPT. Paid plans have been running on GPT-6 Sol since October 7; the free and Go plans have been using GPT-6 Luna since October 8 [A]. Answers can now include charts, buttons, forms and small calculators (OpenAI calls this Intelligent UI). According to OpenAI, it has no separate quota; the plan's existing limits apply [A]. Older desktop apps don't show this view; you need the web version or the current mobile app [A].
Who it affects: Everyone who uses ChatGPT for free. For months we lived with the "free version gets the old model, paid gets the new one" gap; that gap has narrowed but not closed. Sol is still on the paid plan.
What to do: If you're paying for a paid plan, someone on your team already works on a free account, and the job is only short texts, summaries and email drafts, question the paid plan. First, have the same work done on both for two weeks and put the outputs side by side. If you still need the paid plan, the reason is the features that arrived in October, like audio file uploads and Slides; both are on paid plans only, and Slides is on Pro and business plans [A]. Details in our post on GPT-6 in ChatGPT.
What happened: On October 7, Anthropic released Claude Haiku 5.5: $0.10 per million input tokens and $0.50 per million output tokens; the previous model was $1 and $5 [A]. The same day, Sonnet 5.5's cache read price dropped from $0.20 to $0.10 [A]. Three weeks later, on October 26, OpenAI released GPT-6.1 Terra at $1 input and $5 output; the previous Terra was $2 and $12 [A].
Who it affects: Everyone who works with the API or gets automations from their agency. A token is a piece of text the model reads and writes; in Turkish, a single word takes up several tokens. In high-volume work such as customer emails, product descriptions and classification, the bill comes from these two lines.
What to do: In most tools, the model name is a single settings line. Try the small model with 20-30 of your own real emails, and switch if it's good enough. One warning: when you raise the effort setting on Haiku 5.5, the model spends thinking tokens, and those are billed as output tokens; the "cheapest" label may no longer hold once you turn that setting all the way up. For a worked cost example, see the Haiku 5.5 post and the GPT-6.1 Terra post. Our atlas also has automation tools that call these models: n8n and Make.
What happened: Google announced Gemini 4 Argon on September 30 for a limited program open only to cybersecurity teams, then made it generally available on October 13 [A]. In the Gemini app it's available only to Google AI Pro and Ultra subscribers; it isn't on free accounts or on the API's free tier [A]. The API price is $2 input and $10 output per million tokens until November 30; from December 1 it's $4 and $20 [A]. In Turkey, Google AI Pro costs TRY 869.99 (Turkish lira) a month as of October 28, 2026 [A].
Who it affects: People who use it for hard work like complex coding, long agent tasks, and math or science. For writing, summaries and email, paying for this model is unnecessary in my view.
What to do: If you're building an automation with Argon, budget it at $4 and $20. You don't want to build it at the November price and get a surprise in December. On the coding side, if you use Google Antigravity, Argon opened for Pro subscribers with a separate quota that resets weekly [A]. And for Gemini CLI users, Argon support does not appear in the official source as of October 28. The full story is in our Argon post.
What happened: Mistral released the Voxtral 3 family: batch speech-to-text at $0.0035 per minute, a higher-accuracy model at $0.004, and a real-time model at $0.008. According to the official page, Turkish is among the supported languages for both transcription and speech synthesis [A]. The same month, ChatGPT added audio file uploads: you upload a meeting recording and ask for a transcript, a summary and a follow-up email, on paid plans [A].
Who it affects: Every business that has customer conversations, sales calls or meeting recordings. A one-hour recording costs $0.21 on Voxtral's batch model (60 × $0.0035); that calculation is mine, not Mistral's.
What to do: Before you record, think about the other party's consent and your legal obligations (in Turkey, under KVKK, Turkey's Personal Data Protection Law); the price removed the technical barrier, not the legal one. Test the Turkish accuracy with one of your own recordings; the language list on a company's page is not a guarantee of quality. In our atlas, the tools that do this job are Mistral AI Studio and the open-source Whisper.
What happened: On October 6, Google made Nano Banana 2.1 generally available. Text in images, character consistency across edits, and wide aspect ratios up to 8:1 all improved. In the API, a 1K image costs $0.0336, half the price of the previous version; a 4K image costs $0.113 [A].
Who it affects: People producing posters, social media visuals and product scenes. If you use Canva and Midjourney, your alternative just got cheaper.
What to do: How accurately it writes Turkish characters is not clear in the official source. Before putting it on a poster that goes to a client, try it three times with your own text; the letters "ğ", "ş" and "İ" are where it most often breaks. The detailed cost breakdown is in the Nano Banana 2.1 post.
Claude Duet. Anthropic introduced it on October 21 as an "assistant that works without being asked," but it's a research preview, and as of November 1, 2026, there's a waitlist for Pro and Max subscribers in the US [C]. When it will open in Turkey is unknown, and I'm not speculating.
Meta Muse Wonder. Meta announced it on October 12; as of November 1, 2026, the model is available to developers and Muse Code subscribers [A]. There is no official announcement that it has reached Meta AI in WhatsApp and Instagram. So be cautious with "Meta's new model is on WhatsApp" headlines. Meta Muse is in our atlas with a status note.
Anthropic's IPO filing. An S-1 was filed on October 26 [A]. It's big news, but nothing in it changes the price, limits or features of the Claude you use.
DeepSeek V4.1 Flash 1017. Number one on Hugging Face's weekly trending list on October 26 [C]. It's a cheap model: $0.30 per million input tokens during peak hours [A]. But you need to read the privacy terms before sending customer data, and its Turkish performance is not clear in the official source. Our atlas has a DeepSeek page.
| Metric | October 2026 | Note |
|---|---|---|
| New models or model families in the official announcements the guide tracks | 8 | Nano Banana 2.1, Haiku 5.5, GPT-6 (Sol and Luna), Meta Muse Wonder, Gemini 4 Argon, DeepSeek V4.1 Flash 1017, Voxtral 3, GPT-6.1 Terra |
| Models with a price cut | 4 | Haiku 5.5, Sonnet 5.5 cache reads, GPT-6.1 Terra, Nano Banana 2.1 |
| Models with an announced price increase | 1 | Gemini 4 Argon API, December 1 |
| Shut down or retired | 5 | 4 deprecation announcements in the Gemini API, plus an Antigravity preview model shut down on October 5 [A] |
| Tools added to the atlas | 274 | The atlas launched in October; every entry published by the end of October was added that month |
| Most talked-about tool (with source) | Cloudflare security-audit-skill | +15,072 stars in 7 days on GitHub's weekly list, October 26 [C] |
| Biggest download growth in the atlas | Chrome DevTools MCP | npm weekly downloads 1,060,051 → 2,950,197 (178%), October 12-28 [C] |
Star and download counts are measurements, not quality. The question is: does this number matter for a small business's work? The rise of Chrome DevTools MCP matters to developers; for you, it's information.
| Tool | Old | New | Date | Source |
|---|---|---|---|---|
| Claude Haiku 5.5 (per million tokens, input / output) | $1 / $5 (Haiku 4.5) | $0.10 / $0.50 | October 7 | [A] Anthropic |
| Claude Sonnet 5.5 cache reads (per million tokens) | $0.20 | $0.10 | October 7 | [A] Claude pricing page |
| GPT-6.1 Terra (per million tokens, input / output) | $2 / $12 (previous Terra) | $1 / $5 | October 26 | [A] OpenAI pricing page |
| Nano Banana 2.1 (1K image) | twice this in the previous version | $0.0336 | October 6 | [A] Gemini API |
| Gemini 4 Argon API (per million tokens, input / output) | $2 / $10 | $4 / $20 | December 1, 2026 | [A] Gemini API pricing page |
| Gemini 3.1 Pro, 3.6 Flash, 3.7 Flash in Antigravity | Selectable | Removed | October 16 | [A] Google developer forum reply |
| Storage of Gemini API production calls | Off | On by default in new paid projects | October 26 | [A] Gemini API changelog |
The last row isn't about price, but in my view it's the easiest row in the table to miss. Existing projects keep their setting, and it can be turned off per request or in the Google AI Studio settings [A]. If you're opening a new project that processes customer data, check that setting on day one.
For short texts, summaries and email drafts, yes for most work; the upgrade that came with GPT-6 Luna shows that. October features like audio file uploads and Slides, however, are on paid plans. Before deciding, do the same work on both plans for two weeks.
If your current tool works, there's no need to switch. If you have high-volume, simple work (classification, short answers), try Haiku 5.5 or GPT-6.1 Terra with your own examples; if they're good enough, your costs drop significantly. For hard work, calculate models like Argon with the paid plan and the December price.
I won't say "without problems" for any of them. For Voxtral 3, Turkish is on the official language list. For Haiku 5.5, GPT-6.1 Terra and Nano Banana 2.1, Turkish performance is not clear in the official source; you need to test each one with your own text.
Let's work out together which tool is useful for your business and how to bring your bill down: check the AI consulting and AI automation pages, or go straight to START YOUR FREE ANALYSIS .
AI Guide
ChatGPTAnswering questions, drafting emails and other text, summarizing documents, writing product descriptions and social media copy; limited deep research on the free plan.
ClaudeReading, summarizing and comparing long documents, sourced answers with web search, report and text drafts; the multi-step Research feature on paid plans.
GeminiChat, writing, image generation and sourced research reports with Deep Research; use inside Gmail and Docs on paid plans.
Mistral AI Studio (La Plateforme)API access to French company Mistral AI's models (Mistral Large, Medium, Small, Codestral); in the console you create keys, try models and monitor your usage.
Google AI StudioGoogle's developer workspace where you try Gemini models in the browser, draft prompts and apps, and get a free Gemini API key.
Google AntigravityGoogle's agent-based development environment: through a desktop app (Antigravity 2.0), a terminal tool (agy) and IDE extensions, it writes code, edits files, runs commands and uses the browser; it imports Gemini CLI settings.
DeepSeekChat, writing, code and questions that need reasoning; the app is offered free and has a web search mode.
Meta Muse: API and Muse CodeAn API that gives pay-as-you-go access to Meta's Muse models (Muse Wonder, Muse Spark, Muse Image, speech transcription, image segmentation), plus Muse Code, a terminal coding agent that runs on the same models. Because the API is OpenAI- and Anthropic-compatible, existing SDKs and agents connect by changing only the URL and key.Beyond GPT-6, ChatGPT got five new features in October 2026. Audio uploads are on paid plans, Slides on Pro and business plans, locked chats for everyone; Finances is US-only.
+2OpenAI released GPT-6.1 Terra on October 26, 2026: on the API, one million input tokens cost $1 and output tokens $5, half the price of its predecessor. It's available in ChatGPT Work and Codex on Plus and higher plans.
+1Google has deprecated Gemini 2.5 Pro and Flash; starting December 1, Gemini API requests are quietly redirected to 3.8 Flash and 3.5 Flash-Lite. What happens to price and quality, and which automations are affected?
+2We use cookies to improve the site and measure advertising performance. Analytics and advertising cookies run only if you accept them. Details are in our Privacy Policy and Data Protection Notice.
