September 2026 brought two new flagship models, OpenAI’s GPT-6 Astra and Anthropic’s Claude Opus 5.5, plus faster and cheaper tiers from almost every major lab. The main trend among new AI models in September 2026 was lower prices: OpenAI, Anthropic and DeepSeek all cut API rates, and xAI kept Grok 4.7 at its predecessor’s price.
Below you’ll find a dated list of the releases that mattered, what changed in pricing and limits, and what it means for your daily work. Each claim is backed by the company’s own announcement, listed under Sources.
Last updated: October 2026
New AI Models in September 2026 at a Glance
If you missed the summer releases, our July roundup covers them. Here is September in one table:
| Date | Company | Release | What it’s for |
|---|---|---|---|
| Sep 2 | Gemini 3.8 Flash | Fast, low-cost reasoning and coding | |
| Sep 2 | Meta | Muse Spark 1.3 | Agentic and coding work for developers |
| Early Sep | OpenAI | GPT-6 Astra | OpenAI’s most capable model |
| Sep 8 | Meta | Muse (personal agent) | A consumer agent that carries out tasks for you |
| Sep 10 | DeepSeek | DeepSeek-V4.1-Flash | Cheaper multimodal model that replaces V4-Pro in the API |
| Sep 15 | Gemini 3.8 Live and Live Extended Thinking | Real-time voice conversations | |
| Sep 21 | xAI | Grok 4.7 | Coding and knowledge work |
| Sep 22 | Anthropic | Claude Opus 5.5 | Demanding writing, coding and analysis |
| Sep 22 | OpenAI | GPT-6 Sol and GPT-6 Luna | Cheaper, faster GPT-6 tiers |
| Sep 23 | Gemini 3.8 Flash TTS and Flash-Lite TTS | Text-to-speech and custom voices |
The Biggest Model Releases This Month
Here is a short summary of each major launch. For deeper profiles of each model family, see our AI models guide.
OpenAI: GPT-6 Astra, then Sol and Luna
OpenAI introduced GPT-6 Astra early in the month and calls it its most capable model. It is built for complex reasoning, coding, research and computer use, meaning it can operate software and websites for you. The API lists a context window of about 1.05 million tokens. On September 22, OpenAI added GPT-6 Sol and GPT-6 Luna. They are trained with similar methods to Astra but are faster and cheaper, for everyday tasks.
Anthropic: Claude Opus 5.5
Released on September 22, Opus 5.5 is the first model in Anthropic’s Claude 5.5 family. Anthropic says it performs at the level of Claude Fable 5.1 on most work and costs 40% less to run than Opus 5 on typical workloads. It also scored best of any model Anthropic has tested on its automated behavioral audit. Anthropic’s own benchmarks and audit results have not been independently verified.
xAI: Grok 4.7
Grok 4.7 arrived on September 21. xAI calls it its most capable model for coding and knowledge work. It is built on a larger base model than Grok 4.6 and trained to check its own work more carefully. It costs the same as Grok 4.6.
Google: Gemini 3.8 Flash and new voice models
Google released Gemini 3.8 Flash on September 2. It calls this Flash its best reasoning and coding model yet, at the same speed and price as 3.7 Flash. A separate cybersecurity version is available only to vetted defenders. Later in the month, Google added Gemini 3.8 Live models for voice conversations in the Gemini app, Workspace and Search, plus new text-to-speech models.
DeepSeek and Meta
DeepSeek-V4.1-Flash launched on September 10 with built-in image understanding and lower API prices. DeepSeek says it outperforms its own former flagship, V4-Pro. Meta released Muse Spark 1.3 for developers on September 2 and launched Muse on September 8. Muse is a personal agent that runs on Muse Spark and takes tasks off your plate. You use it in the Muse app or in WhatsApp. [VERIFY: Muse availability in GCC countries]
Updates Worth Knowing (Pricing, Limits, Features)
The table shows API prices, which developers pay per million tokens. A token is a small piece of a word, so 1,000 tokens is roughly 750 English words. Consumer apps charge a flat subscription instead, but API prices are a good sign of where costs are heading.
| Model | Input / output per 1M tokens | Change |
|---|---|---|
| Claude Opus 5.5 | $4 / $20 | 20% lower than Opus 5 |
| GPT-6 Astra | $10 / $50 | New model |
| GPT-6 Sol | $2 / $10 | 50% lower than GPT-5.6 Sol’s promotional price |
| GPT-6 Luna | $0.10 / $0.50 | 50% lower than GPT-5.6 Luna’s promotional price |
| Grok 4.7 | $2 / $6 | Unchanged from Grok 4.6 |
| Gemini 3.8 Flash | $0.75 / $3.75 | Same introductory price as 3.7 Flash |
Prices checked on September 29, 2026 from each provider’s official announcement or documentation. [VERIFY PRICE]
Other changes to note:
- DeepSeek model routing: since 04:00 UTC on September 14 (8:00 a.m. in the UAE), requests to DeepSeek’s V4-Pro are answered by V4.1-Flash and billed at Flash rates. This lasts until V4.1-Pro launches. DeepSeek’s off-peak API rates are half its peak rates.
- Claude usage limits: alongside Opus 5.5, Anthropic raised the five-hour usage limits on its paid plans. [VERIFY: confirm on Anthropic’s official pages]
- Gated cyber models: Google (Gemini 3.8 Flash Cyber) and Anthropic (Opus 5.5) now limit some security and biology capabilities to vetted users.
What Changed Inside Lexika This Month
[LEXIKA TEAM TO CONFIRM: which September releases were added to the Lexika gallery (for example GPT-6 Astra, Sol and Luna, Claude Opus 5.5, Grok 4.7, Gemini 3.8 Flash, DeepSeek-V4.1-Flash), their Star costs, and any product changes shipped in September.]
When a new model appears, the Lexika model gallery shows its spec sheet, knowledge cutoff, reasoning support and benchmark scores published by Artificial Analysis. You can see the latest models there and compare them before you use one. Heavier models cost more Stars per message, so check the gallery before you move routine work to a flagship.
What This Means for Your Work
A new release doesn’t mean you have to switch. Here is how to read this month’s news:
- Freelancers paying for several tools: flagship quality is getting cheaper. Mid-tier models such as GPT-6 Sol are aimed at work that needed a top model a few months ago.
- Teams: agents that use your computer or accounts, such as Meta’s Muse or Astra’s computer use, need clear rules on what they can access before anyone connects work systems.
- Students: free and low-cost tiers usually improve after a flagship launch, so wait a few weeks before paying for an upgrade.
Before switching, test the new model on two or three real tasks you already know well, including any Arabic–English work, because published benchmarks rarely measure Arabic quality. Our guide on how to choose an AI model gives you a five-question checklist.
Note: Most benchmark scores in this month’s launches were published by the companies themselves. Treat them as claims until independent results appear.
FAQ
1. What was the biggest AI release of September 2026?
GPT-6 Astra and Claude Opus 5.5 were the two headline launches. Astra is OpenAI’s new top model for complex reasoning, coding and computer use. Opus 5.5 is Anthropic’s new Opus model, which the company says matches Fable 5.1 on most work at a lower cost. Which one matters more depends on your work, so test both on real tasks before deciding.
2. Did AI get cheaper in September 2026?
For developers, yes. OpenAI priced GPT-6 Sol and Luna at half of GPT-5.6’s promotional rates. Anthropic cut Opus 5.5’s token prices by 20% compared with Opus 5, and DeepSeek lowered its API prices. Consumer subscription prices did not change in the same way, but cheaper models often lead to higher usage limits over time.
3. Should I switch to the newest model right away?
Not necessarily. New models can behave differently from the ones you’re used to, and early benchmark claims come mostly from the companies themselves. Run a few of your real tasks through the new model and compare the results with your current one. Switch only if the output is clearly better for your work or costs less for the same quality.
Key takeaways
New AI models in September 2026 raised the ceiling with GPT-6 Astra and Claude Opus 5.5 and lowered prices across the rest of the market. For most people, the practical gain is stronger mid-tier models at lower cost.
- OpenAI and Anthropic both launched new top-tier models in September.
- API prices fell at OpenAI, Anthropic and DeepSeek; xAI held Grok 4.7 at the old price.
- Agents that act for you, such as Meta’s Muse, moved into consumer apps.
- Test new models on your own work before switching.
If you’d rather not juggle three subscriptions to try every new release, you can see the latest models in one gallery and switch between them in a single conversation, with a 7-day free Pro trial.
Sources
- Google: Introducing Gemini 3.8 Flash and 3.8 Flash Cyber (Sep 2, 2026)
- Google: Introducing Gemini 3.8 Live and 3.8 Live Extended Thinking (Sep 15, 2026)
- Meta: Introducing Muse Spark 1.3 (Sep 2, 2026)
- Meta: Introducing Muse (Sep 8, 2026)
- OpenAI: GPT-6 Astra model documentation
- OpenAI: Introducing GPT-6 Sol and Luna (Sep 22, 2026)
- DeepSeek: Introducing DeepSeek-V4.1-Flash (Sep 10, 2026)
- xAI: Introducing Grok 4.7 (Sep 21, 2026)
- Anthropic: Introducing Claude Opus 5.5 (Sep 22, 2026)
