Recent AI news · Original report: 2026-09-30 (KST)

GPT-6.1 Sol price, ChatGPT Live plugins, Claude in AWS Seoul: this week (Sep 30, 2026)

This article was compiled with AI. This English translation preserves the original report's collection date, findings, and limitations. Please consult the linked sources for current information. Originally collected: 2026-09-30 10:10 (KST, UTC+09:00).

Translated into English: 2026-09-30 (KST). The translation date does not indicate a new source check.

At a glance

This week's three announcements come down to price, voice and place. OpenAI said it cut the API price of its new model sharply, ChatGPT's voice conversations can now operate connected apps by speech, and AWS opened a way to run Claude inside its Seoul Region. All performance figures are the announcing companies' own.

This week's announcements

We opened the official documents directly starting at 10:10 a.m. KST on September 30, 2026. Dates are the U.S. dates written in the sources.

OpenAI: GPT-6.1 Sol

OpenAI released GPT-6.1 Sol, an upgrade to GPT-6 Sol. It says the model comes closer to GPT-6 Astra on hard tasks while its standard input and output token prices are one-fifth of Astra's. It also says Astra remains its most capable frontier model overall.

• API price: $2 per million input tokens, $0.10 per million cached input tokens (95% cheaper than standard input) and $10 per million output tokens. The model name is gpt-6.1-sol.
• Where you can use it: Plus, Pro, Business, Enterprise and Edu users can use it in ChatGPT Work and Codex. It is not yet available in Chat.
• Performance figures (OpenAI's own evaluations): on the DeepSWE v1.1 coding benchmark it beats the best GPT-6 Sol score by 6.4 percentage points. Its factual error rate is 4.1%, next to 4.5% for GPT-6 Sol and 4.0% for Astra. The announcement says this test uses hard conversations in which users reported errors and does not represent typical use.
• Ultrafast: Astra Ultrafast and GPT-6.1 Sol Ultrafast are described as up to 8 times faster than Astra. They are available in the API, and on a $500 per month tier they are also available in ChatGPT Work and Codex. The release notes say that tier is called Pro 500, that it is sold only on the web at launch, and that buying credits on Pro 100 or Pro 200 does not unlock it.

OpenAI: ChatGPT Live and plugins

According to the release notes, Live now supports plugins on web, iOS and Android. During a voice conversation you can use the connected apps available to your account and follow the responses as written text in the chat.

• Voice also works in Work. You can ask by speech for documents, presentations, spreadsheets, connected-app actions and browser tasks, and an unfinished task can continue in text after you end the call.
• Free and Go users can use Voice in Chat with the plugins their plan supports. Voice in Work requires both Voice and Work access. Existing app connections, permissions and usage limits still apply.
• The help page says actions that need approval must be approved or declined on screen, and spoken approval is not supported. Live does not support video or screen sharing. Usage is measured over 24 hours and listed as 3 hours for Go (GPT-Live-1 mini) and 3 hours for Plus (GPT-Live-1).

AWS: Claude inside the Seoul Region

According to an AWS blog post dated September 29, Amazon Bedrock now supports in-region inference for Claude Opus 5 and Sonnet 5 in Seoul and Sonnet 5 in Singapore. Requests and data are processed in the Region you call and do not leave it, the post says.

• In exchange, throughput is bounded by that Region's capacity and per-Region quotas apply. Billing follows standard on-demand pricing for the Region you call.
• The post names financial services, healthcare and the public sector as examples of users with local data processing requirements.

What it may mean (our estimate)

• For agent workloads that resend the same context repeatedly, cached-input pricing looks likely to matter a lot in cost calculations.
• Voice looks best suited to handling calendar, email and documents when your hands are busy, but approvals still have to be given on screen.
• Organizations that must process data in Korea have one more option to look at.

What we could not confirm

• Performance and speed figures: all are OpenAI's own evaluation results, and we found no independent verification. We did not check Astra's price itself.
• Not tried by us: we did not confirm availability on Korean accounts, the quality of Korean voice, or which plugins are available.
• AWS Seoul pricing and quotas: the post only says standard pricing, so we did not see actual figures. The models named in the post are Opus 5 and Sonnet 5.
• Announcements we could not open: OpenAI's DevDay 2026 recap and its dots introduction page blocked access, so we left them out.

Sources

• OpenAI, Introducing GPT-6.1 Sol (2026-09-29) https://openai.com/index/introducing-gpt-6-1-sol
• OpenAI Help Center, ChatGPT Release Notes (Sep 29 and Sep 23 entries) https://help.openai.com/en/articles/6825453-chatgpt-release-notes
• OpenAI Help Center, ChatGPT Voice https://help.openai.com/en/articles/20001274-chatgpt-voice
• AWS Blog, Introducing Anthropic models on Amazon Bedrock for in-region inference in Seoul and Singapore (2026-09-29) https://aws.amazon.com/blogs/machine-learning/introducing-anthropic-models-on-amazon-bedrock-for-in-region-inference-in-seoul-and-singapore/