[![AI Newsletter](https://substackcdn.com/image/fetch/$s_!m7md!,w_40,h_40,c_fill,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F41327c80-fe59-416d-aa6f-ab6874177ac7_517x517.png)](https://nlp.elvissaravia.com/)
# [AI Newsletter](https://nlp.elvissaravia.com/)
SubscribeSign in
# 🤖 AI Agents Weekly: Claude Opus 5.5, GPT-6 Sol and Luna, MiMo-V2.6, Step 5 Preview, Google AX, Agensh, and More
### Claude Opus 5.5, GPT-6 Sol and Luna, MiMo-V2.6, Step 5 Preview, Google AX, Agensh, and More
Sep 26, 2026
∙ Paid
19
Share
In today’s issue:
  * Anthropic ships Claude Opus 5.5
  * OpenAI launches GPT-6 Sol and Luna
  * Xiaomi open-sources MiMo-V2.6
  * StepFun previews Step 5
  * Claude agents find a new enzyme
  * Google open-sources AX
  * Prime Intellect launches Sandboxes
  * Agensh scales to 1,024 agents
  * xAI releases Grok 4.7
  * Research agent rewrites its own code


And all the top AI dev news, papers, and tools.
* * *
## Top Stories
### Claude Opus 5.5
[![Claude Opus 5.5 benchmark comparison](https://substackcdn.com/image/fetch/$s_!Dkeo!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F9cd86004-a056-4777-9963-d9d013c360f9_1760x1390.png)](https://substackcdn.com/image/fetch/$s_!Dkeo!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F9cd86004-a056-4777-9963-d9d013c360f9_1760x1390.png)
Anthropic released Claude Opus 5.5, the first model in its Claude 5.5 family. It performs at the level of Claude Fable 5.1 on most tasks and costs about 40% less to run than Opus 5 on typical workloads.
  * **Agentic coding:** 66.4% on Terminal-Bench 4.0 against 55.8% for Fable 5.1 and 57.9% for GPT-6 Astra, plus 54.4% on FrontierCode v1.1 and 57.8% on CursorBench 4.0.
  * **Computer use and knowledge work:** 81.8% on OSWorld 2.0 and a GDPval-AA v2.1 Elo of 1846, the top score in Anthropic’s comparison. GPT-6 Astra still leads on AutomationBench (41.4% against 40.0%) and Terminal-Bench-Science.
  * **Pricing and speed:** $4 input and $20 output per million tokens, down from $5 and $25, with cache reads cut from $0.50 to $0.20. Output is more than 30% faster than Opus 5, and a fast mode offers up to 2.5x speed at $8 and $40.
  * **Alignment:** Anthropic reports its strongest result yet on its automated behavioral audit, and Opus 5.5 attempted to circumvent boundaries about 85% less often than Opus 5.
  * **Availability:** Live as `claude-opus-5-5` on the Claude Platform, AWS, Google Cloud, and Azure, with higher five-hour usage limits on Pro, Max, Team, and seat-based Enterprise plans.


**[Blog](https://www.anthropic.com/claude-opus-5-5)**
* * *
### GPT-6 Sol and Luna
[![GPT-6 Sol and Luna on AutomationBench, score against cost per task](https://substackcdn.com/image/fetch/$s_!cmqp!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F912b5fbb-d2e7-4e1a-9285-b9ad5239ff35_1324x1202.png)](https://substackcdn.com/image/fetch/$s_!cmqp!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F912b5fbb-d2e7-4e1a-9285-b9ad5239ff35_1324x1202.png)
OpenAI added GPT-6 Sol and GPT-6 Luna to the GPT-6 family, training them with methods similar to GPT-6 Astra and cutting API prices by 50% against GPT-5.6 promotional pricing.
  * **Pricing:** Sol drops to $2 input and $10 output per million tokens, and Luna to $0.10 and $0.50.
  * **Business workflows:** On AutomationBench, which tests end-to-end workflows across 47 tools, Sol at xhigh effort scores 33.2% at $0.27 per task, ahead of Claude Opus 5 at max effort (26.9%) at 9% of its cost per task.
  * **Coding:** Sol at max effort scores 68.8% on DeepSWE v1.1, within 1.1 points of Claude Fable 5’s 69.9%, at about 80% lower cost per task. Luna reaches 66.6% at 93% less per task than Opus 5.
  * **Caching for agents:** Prompt caching now gets higher hit rates by default with a 90% discount on cached reads, and changing reasoning effort or toggling tools mid-conversation no longer breaks the cache. GitHub reports over 50% fewer prompt tokens needing fresh processing.
  * **Availability:** In ChatGPT Work and Codex for Plus, Pro, Business, Enterprise, and Edu users, and in the API as `gpt-6-sol` and `gpt-6-luna`.


**[Blog](https://openai.com/index/introducing-gpt-6-sol-and-luna/)**
## This post is for paid subscribers
[Subscribe](https://nlp.elvissaravia.com/subscribe?simple=true&next=https%3A%2F%2Fnlp.elvissaravia.com%2Fp%2Fai-agents-weekly-claude-opus-55-gpt&utm_source=paywall&utm_medium=web&utm_content=217543561)
[Already a paid subscriber? **Sign in**](https://substack.com/sign-in?redirect=%2Fp%2Fai-agents-weekly-claude-opus-55-gpt&for_pub=nlpnews&change_user=false)
© 2026 elvis · [Privacy](https://substack.com/privacy) ∙ [Terms](https://substack.com/tos) ∙ [Collection notice](https://substack.com/ccpa#personal-data-collected)
[ Start your Substack](https://substack.com/signup?utm_source=substack&utm_medium=web&utm_content=footer)[Get the app](https://substack.com/app/app-store-redirect?utm_campaign=app-marketing&utm_content=web-footer-button)
[Substack](https://substack.com) is the home for great culture