AI news, week 40: September 27 - October 3, 2026

👁 115 reads

The September 27 - October 3, 2026 reporting period brought Google’s Gemini 4 Argon announcement, OpenAI’s DevDay launches including GPT-6.1 Sol and Dots, and a $500 ChatGPT Pro tier. Image-generation coverage included Ideogram 4.5, FLUX 3 Image and Qwen-Image-2.1. This is an archive of what the source publication reported that week, with 15 developments, dated price changes and practical account checks. Announced dates and promotions below are historical information, not confirmation that the same offers remain available today.

This is a weekly retrospective. Related daily coverage appears in the AI briefing, commentary in the daily closing report, setbacks in the week in AI failures, generation-price coverage in AI dealsand announcements in the AI calendar. The linked articles provide the underlying sources and test methods. An older announcement is not a substitute for checking the provider’s current status.

15 developments from the reporting period

1. Gemini 4 Argon, September 30. Google announced a one-million-token output limit, compared with the previous 64,000 cited in its announcement, and 13 outright wins across 19 rows in its selected benchmark table. Announced API rates were $2 / $10 per million input/output tokens initially, then $4 / $20. Access at that source check was restricted to Fairwind participants and internal Google teams. See the Gemini 4 Argon announcement, the Gemini 4 guide and Argon vs Astra vs Opus 5.5.
2. OpenAI DevDay 2026, September 29. The source covered more than 20 announcements and attributed a figure of 1.2 billion weekly ChatGPT users to Sam Altman’s opening remarks. GPT-6.1 Sol was announced for the API and Codex. Dots access varied by plan and region: the source described Pro availability as excluding the EEA, with a separate Business Premium route. That does not establish availability for every US or UK account. See DevDay 2026 coverage and OpenAI Dots.
3. GPT-6.1 Sol. The quoted API rates matched GPT-6 Sol at $2 / $10 per million input/output tokens, with cached input reduced from $0.20 to $0.10. In the source author’s Polish writing test, Sol 6.1 scored 85.8/100, compared with 86.4 for Sol 6 and 90.1 for Claude Opus 5.5. Those are close but not identical Sol scores, and they do not measure English writing. The source described access from Plus through supported ChatGPT Work and Codex experiences. See GPT-6.1 Sol pricing and the source test.
4. Reported GPT-6.1 Astra delay, September 28. The source reported that a planned Astra successor was held back following internal tests involving behavior outside task scope and user permissions. It recorded no new date and noted disagreement among reports about whether this was a delay or a cancellation. Treat those details as attributed reporting, rather than a completed public model evaluation. See coverage of the reported GPT-6.1 Astra delay.
5. ChatGPT Pro 500. The announced $500 monthly tier was described as offering 25 times the Plus allowance and an Ultrafast mode of up to 300 tokens per second. The source also reported Pro 200 returning to sale, with its allowance changing from 20 to 10 times Plus from October 30. These are dated plan claims; check the current official entitlement and local tax-inclusive checkout. See ChatGPT Pro 500 and the plan comparison.
6. Claude Sonnet 5.5, September 28. The quoted API rate was unchanged from Sonnet 5 at $2 / $10 per million input/output tokens. The source author’s Polish test scored it at 80.4, 8.5 points above Sonnet 5 and 9.7 below Opus 5.5. It completed 13 tasks in 83 seconds, compared with 188 seconds for Opus in that run. These are one test’s results, not universal latency or English-language scores. See Claude Sonnet 5.5.
7. Claude Opus 5.5 and Sonnet 5.5 in Antigravity, October 3. The source reported access for eligible paid Google AI Pro and Ultra subscribers, excluding trial subscriptions. It also recorded a November 2 removal date for Claude 4.6 from free and AI Plus accounts. Local subscription prices and actual model access should be checked in the account. See Claude 5.5 in Antigravityand the comparison with an Anthropic subscription in Claude through Google AI Pro.
8. Announced free Gemini changes from October 9. The source described free accounts moving to Flash-Lite with a 32,000-token context, Flash moving to AI Plus, and Pro requiring Google AI Pro. Deep Think was described as expanding from Ultra to Pro. These are the changes recorded in this edition, rather than a live account audit. See the free Gemini plan changes.
9. Ideogram 4.5, September 30. The source author compared ten consecutive edits across eight models. Ideogram retained an SSIM of 0.95 in the measured untouched areas, versus 0.58 for Seedream 5.0 Pro. It reproduced 11 of 13 requested Polish lettering examples. The listed price range was $0.008-$0.22 per image. Those results depend on the test images and method. See the Ideogram 4.5 launch, the ten-edit test and the Ideogram guide.
10. FLUX 3 Image, October 1. The source author recorded a 5456 × 3072 output and tested box-guided composition, placing 16 of 16 requested objects in their specified locations in that test. The cited fal.ai introductory price was $0.024 per megapixel through October 8, then twice that rate. This archived promotion should not be treated as a current offer. See the FLUX 3 Image launch, the bounding-box test and the FLUX guide.
11. Qwen-Image-2.1 and open-weight image models, October 2. The cited AA snapshot gave Qwen-Image-2.1 1034 Elo in text-to-image and 1074 in editing: 18th overall and first among the open-weight entries in that comparison. The author’s RTX 4090 run produced a 2048 × 2048 image in 44 seconds and reproduced six of 13 Polish lettering examples. The source also flags commercial-license restrictions; open weights do not by themselves establish permission for every commercial use. Check the actual license for your deployment. See Qwen-Image-2.1 in the AA rankings and the open-weight image model comparison.
12. Eleven v4, September 28. ElevenLabs’ new speech model led the cited AA voice ranking at 1319 points. In the source author’s Polish test, it read 18 of 18 numbers, dates and abbreviations correctly, compared with seven of 18 for Flash v2.5. The listed API rate was $0.08 per 1000 characters, with a $0.022 promotion through October 12. That Polish test does not establish an English voice ranking. See the Eleven v4 language test and the comparison of 14 AI voice models.
13. Kling 4.0 Flash early access, September 28. The source described 720p clips of up to 20 seconds, available through the $159.99 Ultra plan. Full Kling 4.0, advertised with up to 30 seconds and 4K, was expected during October but was not listed on fal.ai or OpenRouter at the source check. See Kling 4.0.
14. Sora shutdown and Veo 3.1 preview retirement. The source records the sora-2 and sora-2-pro API shutdown on September 24, after the app’s earlier closure. It also records three Veo 3.1 preview models in Gemini API’s retirement schedule with October 22 as the earliest date and gemini-omni-1.1-flash as the replacement. Those provider-specific preview entries should not be confused with every Veo access route. See the Sora shutdown, six Sora alternatives and the Veo 3.1 preview retirement notice.
15. Anthropic IPO reporting and a White House agreement. The edition reported an Anthropic investor day planned for October 14 in San Francisco and a prospective Nasdaq listing before November 26. It distinguished a reported $965 billion valuation from investor expectations of $1.8-$2 trillion; plans and expectations are not completed transactions. It also reported a September 30 voluntary AI-safety agreement with the Trump administration involving Jensen Huang, Sundar Pichai, Mark Zuckerberg, Elon Musk, Dario Amodei and Greg Brockman. See the attributed coverage in Anthropic IPO reporting and the White House agreement.

Dated price changes

API figures below remain in the provider’s quoted US dollars. Another country’s retail subscription prices have not been converted into purported US or UK offers. Check local checkout prices, taxes and account eligibility separately.

Subscription changes recorded in this edition

The original edition also compared paid Claude and ChatGPT subscriptions. Their local retail prices are not reproduced as dollar quotes here. For a needs-based comparison, see the AI plan selection guide and AI subscription comparison.

API and credit rates in US dollars

Kimi K3 and DeepSeek figures came from the public OpenRouter API checked on October 2. For other dated image and video pricing coverage, see AI deals. Text-model workload costs are discussed in How much does AI cost?.

Three account checks prompted by these announcements

1. Check whether the available Gemini free model meets your needs. The original edition suggested comparing Flash-Lite on three representative tasks before the announced October 9 change. For a current decision, inspect the models actually available in your account and compare them on your own work before buying a plan. See the Gemini free-tier guide.
2. Inspect Antigravity notices and the signed-in account. If a notice appears beside Claude 4.6, read its full text and check which Google account is signed in. A subscription on a different account may not provide access in the editor. The edition recorded November 2 as the announced removal date for the affected free and AI Plus routes; the current notice is the better source for your account. See the step-by-step account guide.
3. Review Sora exports and Veo preview dependencies. The source directed Sora users to sora.chatgpt.com/sunset for an Export option and recorded a permanent-deletion announcement without an exact date. As an archived instruction, this does not guarantee the export remains available. For code using veo-3.1-generate-preview, review the current retirement notice and test the documented replacement before changing production traffic. See the Veo-to-Omni migration guide.

Related model guides and comparison tools

The original edition linked a new collection of brand guides alongside launch coverage for Gemini 4, ElevenLabs, Kling 4.0, FLUX and Ideogram. The directory below groups the corresponding topics. A guide may contain a hands-on test, a sourced measurement or a manufacturer summary; those forms of evidence should remain clearly distinguished.

The source edition also introduced three comparison resources. Their datasets and market assumptions should be checked before reusing the results:

• Language-model comparison arena - responses from 31 models to the same nine tasks, with evaluator scores. The source tasks were in Polish, not a fresh English benchmark.
• Image-generator comparison arena - 11 models across four tests: Polish lettering, bounding-box composition, ten successive edits and 4K output.
• AI plan selection quiz - seven questions producing one to three suggestions in the source tool. Local price assumptions must be checked before treating a recommendation as applicable to another market.

Related topics to explore

The source publication used its own Search Console data to select these subjects. That traffic is not this site’s audience data, so the list below is presented as related reading rather than a local popularity ranking:

1. The TikTok Hotel Lobby trend: Hotel Lobby AI: workflow and costs,
2. Claude Opus 5.5 compared with GPT-6 Astra: nine tests and their costs,
3. GPT-6.1 Sol: pricing and the source test,
4. Gemini 4: the Argon guide,
5. OpenAI Dots and ChatGPT Pro 500: Dots access and regional limits and Pro 500 pricing.

Dates announced as of the end of this edition

Announcements without a precise date, including full Kling 4.0 and wider Argon access, are collected in the AI calendar and October AI announcements.

How this retrospective was prepared

• Reporting period: Sunday through Saturday, September 27 - October 3. The source labels this the week 40 edition; the date range is retained explicitly because it is not exactly an ISO Monday-to-Sunday week.
• Material: the source publication’s articles from that period, provider model listings and its own search-traffic analysis. Those traffic figures are not transferred to this localized site.
• Numbers are attributed to the linked reporting or its described tests. Announcements, reported plans and completed measurements remain separate.
• API rates retain their quoted currency. Unverified local retail prices are omitted, and expired promotional dates are presented as archive information.

Google · Your sourcesQuantslant higher in your Google - one clickAdd to preferred sources →

Sources and verification date

This adaptation draws on the source publication’s September 27 - October 3, 2026 articles linked above and their cited material: Google and DeepMind, OpenAI DevDay coverage and pricing documentation, Anthropic, Ideogram, Black Forest Labs, Artificial Analysis, ElevenLabs, Kling AI, Gemini API retirement documentation, IPO reporting attributed to the Financial Times and Reuters, and the public OpenRouter API. Described original tests include language-model evaluations and image tests from September 30 - October 2. The source edition was checked on October 3, 2026; this is not a claim that all announcements or promotions were reverified on the localization date.

Kinetyra

The link to ElevenLabs is a partner link: if you buy through it, QUANTSLANT earns a commission and you pay the same price. It has no influence on the content, including where the coverage is critical.