# Claude Fable 5.1 and GPT-6 Astra both land, plus Gemini 3.8 Flash

> Week of 31 August 2026: Anthropic ships Fable 5.1 and Mythos 5.1, OpenAI launches GPT-6 Astra, Google releases Gemini 3.8 Flash, and Arabic models appear.

Published 13 Oct 2026. Coverage: 31 Aug 2026 to 6 Sept 2026. By Ali Asif.

## Items

1. **Claude Fable 5.1 and Claude Mythos 5.1** (Anthropic, Frontier): One underlying model at two safeguard levels, with cache reads 75% cheaper and base prices unchanged at $10 in and $50 out per million tokens. [Primary source](https://www.anthropic.com/claude-fable-and-mythos-5-1)
2. **GPT-6 Astra** (OpenAI, Frontier): OpenAI's flagship for computer use, coding and research, with a 1.05M-token window, 128K output and $10 in and $50 out per million tokens. [Primary source](https://developers.openai.com/api/docs/models/gpt-6-astra)
3. **Gemini 3.8 Flash and Gemini 3.8 Flash Cyber** (Google DeepMind, Frontier): Google's most capable Flash model for long-horizon software and agent work, with a separate cyber variant for defenders. [Primary source](https://blog.google/innovation-and-ai/models-and-research/gemini-models/3-8-flash-and-3-8-flash-cyber/)
4. **Muse Spark 1.3** (Meta, Frontier): Meta's latest Muse Spark checkpoint, tuned for multi-step agentic work and coding, with better coding than 1.2. [Primary source](https://research.meta.ai/blog/introducing-muse-spark-1-3)
5. **K2-Horizon-375B-A23B** (MBZUAI Institute of Foundation Models, Open weights): A 375B sparse model with 23B active parameters and 512K context, Apache 2.0, from the Abu Dhabi team, part of a family that spans 0.9B to 375B. [Primary source](https://huggingface.co/IFM/K2-Horizon-375B-A23B)
6. **HUMAIN M3** (HUMAIN, Frontier): An Arabic-focused 428B multimodal model, in research preview on HUMAIN Node, with Arabic post-training added on a MiniMax-M3 base. [Primary source](https://node.humain.com/)
7. **WeatherNext 3** (Google DeepMind, Science): DeepMind's most advanced global weather model yet, following the cyclone work from last month. [Primary source](https://blog.google/innovation-and-ai/models-and-research/google-deepmind/introducing-weathernext-3/)
8. **TimesFM-3** (Google Research, Science): A 330M-parameter zero-shot forecasting model that now handles several targets and covariates at once, relevant to demand and budget forecasting. [Primary source](https://research.google/blog/timesfm-3-a-zero-shot-foundation-model-for-multivariate-forecasting/)
9. **Can LLMs create and evolve their own agent scaffold?** (arXiv, Agents and coding): The most upvoted paper of the week asks whether a fixed model can improve itself just by rewriting the code that surrounds it. [Primary source](https://arxiv.org/abs/2609.01437)

The two biggest labs both shipped a flagship within three days of each other, and I spent more time on the price tables than on the benchmarks.

## Claude Fable 5.1 and Mythos 5.1

Anthropic released both on 1 September. Fable 5.1 is generally available. Mythos 5.1 is the same model with more permissive safeguards, limited to vetted cybersecurity and life sciences users through trusted-access programs. Anthropic says the safeguards got more precise: cyber interventions in Claude Code drop by about 60 percent per session, and biology safeguards fire 85 percent less often on harmless requests.

The commercial change is in caching. Cache reads fall 75 percent to $0.25 per million tokens, which Anthropic estimates trims typical costs about 25 percent and highly agentic ones by up to about 45 percent. Base rates stay at $10 per million input tokens and $50 per million output. Reported results against Fable 5 include Terminal-Bench 4.0 at 55.8 percent versus 42.0, and Humanity's Last Exam without tools at 60.9 versus 57.8.

## GPT-6 Astra

OpenAI announced GPT-6 Astra on 3 September. The docs describe it as the most capable model in the lineup for complex reasoning, coding, computer use, research and document creation. It has a 1,050,000-token context window, up to 128,000 output tokens and an April 2026 knowledge cutoff, and costs $10 per million input tokens and $50 per million output. Batch and Flex run at half the standard rate.

## Gemini 3.8 Flash and Muse Spark 1.3

Google released Gemini 3.8 Flash and a Cyber variant on 2 September, aimed at long-horizon software engineering and agents, and launched the Fairwind Program for trusted cyber defenders the same day. Meta released Muse Spark 1.3 on 2 September, tuned for multi-step tool and browser work.

## Regional and open models

On 3 September the Institute of Foundation Models at MBZUAI released the K2 Horizon family, from 0.9B to 375B parameters, and the 375B flagship has Apache 2.0 weights and 512K context. HUMAIN also introduced M3, an Arabic-focused 428B multimodal model in limited preview on HUMAIN Node.

## Science and forecasting

DeepMind released WeatherNext 3 on 3 September, and Google Research released TimesFM-3 on 31 August, a 330M-parameter time-series model for several targets at once. A paper asking whether an LLM can rewrite its own agent scaffold topped the Hugging Face list.

## What I'd watch next

Whether other labs match the cache-read price cut, since that is the line item that decides what long agent runs cost.

## My take

Two flagships in three days, and for once I care more about price than score. Fable 5.1 cut cache-read cost by 75 percent, which matters because my agents re-read the same long documents all day. The HUMAIN M3 release is also good news for the Kingdom: an Arabic-first model hosted locally is the kind of option Saudi buyers have been asking for.

## Sources

- https://www.anthropic.com/claude-fable-and-mythos-5-1
- https://developers.openai.com/api/docs/models/gpt-6-astra
- https://blog.google/innovation-and-ai/models-and-research/gemini-models/3-8-flash-and-3-8-flash-cyber/
- https://research.meta.ai/blog/introducing-muse-spark-1-3
- https://huggingface.co/IFM/K2-Horizon-375B-A23B
- https://node.humain.com/
- https://blog.google/innovation-and-ai/models-and-research/google-deepmind/introducing-weathernext-3/
- https://research.google/blog/timesfm-3-a-zero-shot-foundation-model-for-multivariate-forecasting/
- https://arxiv.org/abs/2609.01437

Canonical: https://heyaliasif.com/ai/2026-08-31-week-in-ai