New AI models in 2026: which one should you study with?

Opus 5.5, GPT-6 Sol, Grok 4.7, Gemini 3.8 Flash, Qwen3.8… Between August and September 2026 alone, the big AI labs shipped a dozen or so models. As a student, the question isn’t which one tops a benchmark this week. It’s which one to use, ideally for free, to revise, write, code or find sources. Here’s a fact-checked snapshot as of September 27, 2026, followed by a use-by-use guide.

In short: the three big apps (ChatGPT, Claude, Gemini) all have a solid free tier, but their most powerful models stay behind a paywall. For free, ChatGPT offers unlimited text chats with GPT-5.6 Luna, Claude gives you Sonnet 5 and Haiku 4.5, and Gemini gives you 3.6 Flash plus variable access to 3.1 Pro. Google also offers eligible students aged 18+ a year of Google AI Plus at no cost, and Mistral has a student plan at €5.99 a month before tax. The best choice isn’t one model: it’s a free main tool, a second opinion, and a study method that doesn’t depend on either.

Three ideas to keep things straight

• A model isn’t an app. The model (Opus 5.5, GPT-6 Sol…) is the engine. The app (Claude, ChatGPT, Gemini…) is the car, and your plan decides which engine you get.
• API prices aren’t subscription prices. “Per million tokens” prices are for developers. You pay a monthly subscription, or nothing.
• Open weights. Some labs (Mistral, DeepSeek, Qwen) publish their model weights: anyone can download and run them, sometimes on their own machine for the smaller ones.

Anthropic: Claude

Anthropic’s documentation lists four current models:

• Fable 5.1: for demanding reasoning and long-horizon work, the most powerful and most expensive ($10 / $50 per million input / output tokens). Its twin, Mythos 5.1, which has different safeguards, is limited to trusted access programs in cybersecurity and the life sciences.
• Opus 5.5: released September 22, for long-running coding and knowledge work. Anthropic says it performs at the level of Fable 5.1 on most work, for $4 / $20.
• Sonnet 5: the best balance of speed and intelligence, per Anthropic, at $2 / $10.
• Haiku 4.5: the fastest, at $1 / $5.

In the app, according to the plans page: free with Sonnet and Haiku (no Opus), Pro at $17 a month billed annually or $20 monthly with Opus, and Max from $100. All prices exclude tax.

OpenAI: ChatGPT

On September 3, OpenAI introduced GPT-6 Astra, its most powerful model, available to Plus, Pro, Business and Enterprise subscribers. On September 22 came GPT-6 Sol and GPT-6 Luna: Sol for demanding everyday work, Luna for speed and volume, with API prices cut in half compared with their GPT-5.6 equivalents. According to OpenAI, they’re available in the Work and Codex tabs for subscribers, and Free and Go users can access GPT-6 Luna in the desktop app, though not yet in Chat.

On plans, ChatGPT’s pricing page lists unlimited text chats with GPT-5.6 Luna on the free tier (subject to abuse guardrails), plus a Study mode on every plan, free included. Go costs $8 a month in the US (localized pricing in some markets), Plus is $20 a month, and Pro starts at $100. New sign-ups for the $200 Pro plan have been paused since September 10.

Google: Gemini

Google runs two lines: the Pro models (Gemini 3.1 Pro) and the Flash models, which change almost monthly. Gemini 3.8 Flash, released September 2, is reserved in the app for Google AI Pro and Ultra subscribers.

The subscriptions page, as displayed from France, lists:

• Free: 3.6 Flash, variable access to 3.1 Pro, Deep Research, Gemini Live, Canvas;
• Google AI Plus: €4.99 a month, double the limits;
• Google AI Pro: €21.99 a month;
• Google AI Ultra: from €99.99 a month.

For students, Google offers one year of Google AI Plus at no cost to eligible students aged 18 and over, to redeem by December 31, 2026 (Google AI Pro in the US). A payment method is required, and the plan renews at the standard price unless you cancel. The app also offers study notebooks built from your own course materials, with quizzes and flashcards. According to Google, the Gemini app passed 1 billion monthly users in August.

Mistral: Vibe (formerly Le Chat)

A name change worth knowing: “Le Chat is now Vibe,” Mistral announced in late May. Vibe combines the everyday assistant and the coding agent. Its flagship model, Mistral Medium 3.5, is released as open weights under a modified MIT license.

According to its pricing page, Vibe has a free tier (limited messages), a Pro plan at $14.99 / €14.99 a month before tax, and an Education plan at $5.99 / €5.99 a month before tax, valid for up to 12 months for verified students at accredited higher-education institutions who haven’t used Vibe or Le Chat before. On September 8, the French company raised €3 billion, which it calls the largest equity round ever completed by a European tech company.

DeepSeek

On September 10, the Chinese lab released DeepSeek-V4.1-Flash, a model with native image understanding, as open weights on Hugging Face. DeepSeek says it outperforms its previous flagship, V4-Pro, which is being phased out until a V4.1-Pro arrives. The web chat and app are offered with no subscription listed on the site: only the API is billed. For questions about data and hosting, see our article on Chinese AI in Europe.

Qwen (Alibaba)

In August, Qwen released the Qwen3.8 family, including, for the first time, a “Max”-class model with open weights: Qwen3.8-2.4T-A95B, with 2.4 trillion parameters of which 95 billion are active, under a Qwen-specific license. Much smaller versions exist (27 billion parameters). You can use it online through Qwen Chat. We couldn’t find a verifiable consumer price list for the app.

Meta: Muse

On September 8, Meta launched Muse, a personal agent that acts on your behalf (booking, filling in forms, sending emails once you approve), powered by Muse Spark, which Meta calls its most capable model to date. Muse is rolling out in the US only for now. Meta says it’s “free for most of what people need,” with subscription plans for more.

SpaceXAI: Grok

xAI now presents itself as SpaceXAI on its website. Grok 4.7, released September 21, is its most capable model for coding and knowledge work, available in the Cursor editor, Grok Build and the API. For the app, the pricing page shows a free tier, SuperGrok at $30 a month and SuperGrok Plus at $100.

The recap

App	Flagship (Sept. 2026)	What the free tier gives you	First paid plan
Claude	Fable 5.1, Opus 5.5	Sonnet 5, Haiku 4.5	Pro: $17–20 / month
ChatGPT	GPT-6 Astra	Unlimited GPT-5.6 Luna (text), Study mode	Go: $8 / month (US)
Gemini	3.1 Pro, 3.8 Flash	3.6 Flash, variable 3.1 Pro	AI Plus: €4.99 / month
Vibe (Mistral)	Medium 3.5	Limited messages	Education: €5.99 / month + tax
DeepSeek	V4.1-Flash	Chat and app	API only
Grok	Grok 4.7	Limited usage	SuperGrok: $30 / month

Which model for which student use

What follows is our take, based on the features each company announces, not on an independent test.

Revising (quizzes, summary sheets, planning). Free Gemini is well equipped with study notebooks and quizzes built from your notes. ChatGPT’s Study mode, also free, guides you with questions instead of handing you answers. Claude Sonnet 5 does the job well with a clear instruction: “quiz me, don’t give me the answer.”

Writing with a coach. Any of the big three works, but always on your own text: ask for feedback on structure, clarity and transitions, never a full rewrite. According to Anthropic, text from recent Claude models carries an invisible watermark tied to the EU AI Act: one more reason to stay the author.

Coding. Claude (free Sonnet, Opus on Pro), GPT-6 Sol in Codex for subscribers, Grok 4.7 in Cursor. To learn without a subscription or work offline, open-weight models (Mistral, DeepSeek, Qwen) are a real option.

Math and science. Every major model can break down a calculation step by step. The right move: submit your own working and ask where it breaks. Always redo the calculation yourself; an AI math error usually comes nicely formatted.

Languages. Voice modes make a real difference for speaking practice: Gemini Live is in the free tier, and the free tiers of ChatGPT and Claude include voice too (limited on ChatGPT). Ask to be corrected on grammar and pronunciation, in the target language.

Research. Deep Research is included in free Gemini, limited in free ChatGPT, and paid-only on Claude. Either way, open every source it cites: invented references are still the number one trap.

The point of all this: depend on none of them

In a single week of September, Anthropic, OpenAI and SpaceXAI each shipped a new model, and prices moved. Betting your studies on one tool means your method breaks at the next plan change. A few simple principles:

1. One free main tool, one second opinion. When an answer matters (a key concept before an exam, a source), ask a second model. If they disagree, that’s your cue to check your course.
2. Keep your prompts and notes on your side. Your study instructions, summary sheets and schedule belong in your own files, not just in an app’s history. They’ll work anywhere.
3. Know the open option. Knowing an open-weight model can take over protects you from a price hike or an outage.
4. Never paste personal or confidential data, yours or anyone else’s.
5. Check your school’s rules on AI use, course by course.

What it concretely changes for you: you can do almost everything for free in 2026, as long as you use AI as a coach that makes you work, not as a shortcut. Subscriptions pay off for heavy coding or in-depth research, and the student offers from Google and Mistral cut the bill.

FAQ

What’s the best free AI for studying in 2026?

There isn’t a single one. Free Gemini is very complete for revision (quizzes, notebooks, Deep Research), free ChatGPT offers unlimited text chats and a Study mode, and free Claude gives you Sonnet 5. Master one and keep another for double-checking.

Are there student discounts on AI subscriptions?

Yes, as of September 27, 2026: Google offers a year of Google AI Plus to eligible students aged 18+ (redeem by December 31, 2026), and Mistral offers an Education plan at €5.99 a month before tax for up to 12 months. Check the terms before signing up.

Do I need to pay to get the best model?

For the most powerful models (Opus 5.5, GPT-6 Astra, Gemini 3.8 Flash in the app), yes. But for revising, explaining a lecture or checking your reasoning, free models are enough in the vast majority of cases.

Can I use Chinese models like DeepSeek or Qwen?

Their models are often released as open weights, so they can run without going through their servers. Using their online apps raises other questions, especially about data: our article on Chinese AI in Europe covers it.

Further reading

• Claude Opus 5.5: what really changes: pricing, effort levels and benchmarks for Anthropic’s latest model.
• Studying with AI without cheating: how to keep AI in a coaching role.
• The Revision sheet skill: turns your course into a sheet you fill in yourself.
• The Scientific articles connector: find real sources and cite them correctly.