AI at the End of 2026 – What Changed in a Year?
Published
In 2026, all three major model developers – Anthropic, OpenAI and Google – released a new generation of models. At the same time, the focus moved from conversation to action: the newest models use a computer, write and run code, and carry out multi-step tasks from start to finish. This post sums up what happened during the year and what it means for everyday users.
The year's most important model releases
Anthropic: Claude 5 and 5.5
- June: Claude Fable 5 and Claude Mythos 5 – a new model tier above Opus.
- 24 July: Claude Opus 5.
- September: Fable 5.1 (1 Sep), Opus 5.5 (22 Sep) and Sonnet 5.5 (28 Sep). According to Anthropic, Opus 5.5 performs at the level of Fable 5.1 on most work and costs 40% less to run than Opus 5 on typical workloads.
- 9 November: Fable 5.2 and Mythos 5.2. Mythos 5.2 has the same capabilities as Fable 5.2 but is offered only to organisations verified by Anthropic, for example through its cyber verification programme.
OpenAI: GPT-6
- 3 September: GPT-6 Astra. OpenAI highlighted computer use, coding, cybersecurity and science.
- 22 September: the faster and cheaper GPT-6 Sol and GPT-6 Luna.
- 7 October: GPT-6 for all ChatGPT users. The new Intelligent UI can answer with visuals and interactive elements as well as text. Paid tiers use Sol, free tiers use Luna.
- 17 November: GPT-6.2 Astra and Sol in preview in ChatGPT Work and Codex.
Google: Gemini 4
- 30 September: Gemini 4 Argon was announced and first given to a limited group of trusted cyber defenders.
- 13 October: Argon opened to developers, enterprises and Google AI Pro and Ultra subscribers. In October Google also announced Gemini 4 Carbon.
- November: the fast Gemini 4 Hydrogen, which brings Gemini 4 to every tier and to Google Search.
Three trends shared by everyone
1. AI does things, not just answers
OpenAI markets Astra as a model that can handle almost any work done on a computer. Google says it built Argon for long, multi-step workflows. In practice, people increasingly hand AI a whole task instead of a single question. More on this in AI Agents at Home and at Work in 2026.
2. The most capable models are released in stages
The most capable models no longer open to everyone on the same day. Google gave Argon to trusted cyber defenders first, and Anthropic offers Mythos 5.2 only to verified organisations. OpenAI says it treats its October GPT-6 models as high capability in cybersecurity and in biology and chemistry. Google explains that releasing frontier capabilities at this level safely requires a phased approach.
3. The same capability for less money
Prices fell even as models improved. Fable 5.2 costs 8 dollars per million input tokens and 40 dollars per million output tokens in the API, while Fable 5.1 cost 10 and 50 dollars. OpenAI cut API prices for Sol and Luna to half of GPT-5.6's promotional pricing. For consumers, this shows in the fact that even free versions now run a new-generation model.
What about Finland?
- According to Statistics Finland, the share of 16–89-year-olds who had used generative AI in the previous three months rose from 23% to 41% in 2025. The most common use was searching for information.
- 38% of Finnish enterprises used AI technologies in spring 2025, up from 24% a year earlier.
- In a consumer survey in February 2026, 46% said they did not use AI at all yet. Even so, use in Finland was slightly more common than the EU average, and four out of five said AI speeds up their work.
Regulation moved forward too: the transparency obligations of the EU AI Act took effect in August, and the timeline for high-risk systems was postponed. Read more in The EU AI Act in 2026.
What does this mean for you?
- Try again. If your last attempt was a year ago, you will now be using an entirely new generation of models.
- Give a task, not just a question. Say what you want finished and for whom. Tips in Prompting in 2026.
- Check important facts. Even a better model can be wrong, especially with numbers and recent events.
- Think about what you share. Don't enter passwords or sensitive personal data into a service whose terms you don't know.
Conclusion
2026 did not bring one single breakthrough but a steady flow of better and cheaper models that can do more on their own. That is why it pays to keep up your own AI literacy: knowing what AI can do, where it makes mistakes and when a result needs checking.
Sources
- Anthropic: Models (2026 releases)
- Anthropic: Claude Opus 5.5
- Anthropic: Claude Platform release notes (Fable 5.2, pricing)
- OpenAI: 2026 releases (GPT-6 Astra, Sol, Luna, GPT-6.2)
- OpenAI: GPT-6 and Intelligent UI for everyone
- OpenAI: GPT-6 Sol and GPT-6 Luna, October 2026 update
- OpenAI: Announcing GPT-6 Sol and GPT-6 Luna (pricing)
- Google: Gemini 4 Argon (30 Sep 2026)
- Google: Gemini 4 Argon generally available (13 Oct 2026)
- Google: AI news October 2026 (Gemini 4 Carbon)
- Google: Introducing Gemini 4 Hydrogen
- Statistics Finland: use of generative AI rose to 41% (in Finnish)
- Statistics Finland: ICT use in enterprises 2025 (in Finnish)
- Statistics Finland Tietotrendit: most working people use AI (20 May 2026, in Finnish)