September 2, 2026 is turning into one of the biggest days in AI for this year.
Today, Google DeepMind released Google Gemini 3.8 Flash, the most powerful Flash model yet. Within the same day, also a big update: Anthropic released Claude Fable 5.1 and Claude Mythos 5.1.
Google Gemini 3.8 Flash — What Is New?
Google DeepMind has launched Gemini 3.8 Flash just three weeks after it launched Gemini 3.7 Flash on August 13. That makes Gemini 3.8 Flash the third update to Google’s workhorse Flash models in about six weeks, one of the fastest release cycles to reach production for any product by any large organization.
The new Flash model was codenamed “Skimaki” internally by Google DeepMind, and had been revealed to be under development by Google, according to a Wall Street Journal report, to help close the coding abilities gap with other top tier providers such as OpenAI and Anthropic.
What Gemini 3.8 Flash Improves
Here is a summary of what Google says has improved in this workhorse model:
- Software engineering — Gemini 3.8 Flash will be much smarter at writing and debugging code in software engineering tasks.
- Agentic tasks — Gemini 3.8 is able to manage multi-step autonomous workflows better.
- Multi-step reasoning — Much stronger multi-step reasoning for a variety of specialized problems.
- Long-horizon coding — Gemini 3.8 can follow large, complex codebases.
Gemini 3.8 Flash Pricing
| Tier | Price |
|---|---|
| Input tokens | $0.75 per million |
| Output tokens | $3.75 per million |
| Context window | 1,048,576 tokens |
Same price as 3.7 Flash — no cost increase for a meaningfully smarter model.
Where It Is Available
Gemini 3.8 Flash is live today in the Gemini app for Google AI Pro and Ultra subscribers, in AI Mode, in Gemini in Google Sheets, in Google AI Studio, and as part of the Gemini API.
Gemini 3.8 Flash Cyber
Google has introduced a security-specific model Gemini 3.8 Flash Cyber, replacing the previous Gemini 3.5 Cyber model. Gemini 3.8 Flash Cyber is an autonomous vulnerability discovery and patching model that scores 47.2% on the CWE-Bench patching benchmark, comparable to leading frontier language models at lower cost. It will only be available through Google’s Fairwind Program for government, critical infrastructure, and security researchers and companies.
Claude Fable 5.1 — Anthropic’s Biggest Coding Leap Yet
In the meantime, Google’s rival Anthropic released an update of their model Claude, now called Claude Fable 5.1. Although Google received more attention for the update of their search engine model Gemini to Gemini 3.8 Flash, the increase in benchmark scores of Fable 5.1 is even more impressive.
The Benchmark Numbers
Terminal-bench Science (a very scientific reasoning benchmark): 52.6% (up from 24.7% for Fable 5). This is more than a 100% increase in a single pass at updating the model’s weights.
Additionally, on WebArena 2.0 (simulating use of a computer to complete real world tasks), the score rises to 77.9% (up from 72.9% for Fable 5).
The model attains a score of 73.4% on the Cursor Bench (coding agent performance), up from 70.5% on the previous version of Fable.
What Fable 5.1 Does Better
- Coding agent performance — significantly stronger on real-world programming tasks.
- Scientific research — scored more than double the previous score on scientific reasoning.
- Computer use — operation of software and completion of complex multi-step tasks.
- Root cause debugging — improved ability to identify and explain why a piece of code is not working.
- Business automation — improved performance on enterprise workflow tasks.
- Agentic reasoning — improved in working to achieve a complex objective autonomously.
This AI can take a complex research goal and implement it by writing code, running experiments, verifying results, identifying problems, and continuing to work independently to solve them for extended periods of time. Anthropic calls this a genuine autonomous AI research assistant.
Claude Mythos 5.1 — The Restricted Powerhouse
Anthropic has also today launched Claude Mythos 5.1. This model has been restricted in a very different way to Fable 5.1, and is intended for use by highly advanced professionals working in only two fields: cybersecurity and life sciences research.
Like Fable 5.1, Claude Mythos 5.1 is based on the same underlying model, but with a different set of safeguards and restrictions to access and modify parameters. Mythos is a different class of model aimed at the highest-end professional work in a pair of realms: cybersecurity, and life sciences research.
What Mythos 5.1 Can Do
Testing from Anthropic was quite impressive. Demonstration of life sciences research had the system generating protein binders with nearly 50% hit rate for 12 different research targets. In addition to this work in life sciences research, Mythos 5.1 also generated GPU optimization to several open-source biology models — increasing the speed of seven of them by up to 2.5 times.
In short, Anthropic is testing out a new AI scientific researcher that can autonomously perform a lot of the work of a biologist or drug discovery researcher.
The Bigger Picture — AI Is Moving Faster Than Ever
The launch today by two of the leading AI organizations of very major updates to their leading models marks another huge step in the very fast developing AI race.
These announcements close a coding gap (Gemini 3.8 Flash), more than double the scientific reasoning of the world’s leading general purpose scientific AI (Claude Fable 5.1), and lay the foundations of what amounts to AI-assisted scientific research (Claude Mythos 5.1).
The race for the development of AI models is not slowing down. The companies that will benefit most from these developments are those already using these models for content, research, development, and automation.
Quick Comparison — What Launched Today
| Model | Company | Best For | Available |
|---|---|---|---|
| Gemini 3.8 Flash | Coding, agents, reasoning | Public ✅ | |
| Gemini 3.8 Flash Cyber | Cybersecurity, vulnerability patching | Fairwind Program only | |
| Claude Fable 5.1 | Anthropic | Coding, science, computer use | Public ✅ |
| Claude Mythos 5.1 | Anthropic | Cybersecurity, life sciences research | Restricted access |
The Bottom Line
If you use AI tools to write, code, or research, today is a significant day. Google closed a major coding gap with Gemini 3.8 Flash. Anthropic more than doubled the scientific reasoning performance of Fable 5.1. And Mythos 5.1 hints at a near future where AI does not just assist in scientific research — it conducts it autonomously.
The AI race in 2026 is moving faster than ever — and the people and companies using these tools today are the ones who will benefit most tomorrow.
