2026-09-26 notis
OTHER NOTIS SITES AIYOU ARE HERE AI models, prices and agents. ainotis.com Hiking Trail openings, hut bookings, and route conditions. hiking.notis.news Photography Cameras, lenses and firmware. photo.notis.news More areas when we can cover them every day.
CATEGORY Join
Show Read as Find Sections Models and releasesAgents and appsResearchMoney and computeLaws and rulingsSafety and securityAI at work More The monthlyArchiveModelsAboutCorrectionsThe numbersRSS
My notis
TODAY'S NEWS
September 2026 5 STORIES
MTWTFSS
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30

Days with a dot have news. See the archive.

The whole week ›
15:46
0/5
WEATHER THEME CARD SPACING

Today's news

today

My page

Kept in this browser only.

SAVED FOLLOWING COLLECTIONS READING SINCE
SAVED STORIES
Nothing saved yet. Tap the bookmark on a card to keep it here. read today
CASES YOU FOLLOW all cases

No case followed yet.

COLLECTIONS
Price changes Stories where a published price went up or down.
Models being retired Models being withdrawn, and gaps before a replacement is ready.
THE MORNING EMAIL The day's top stories. Sign up
toned down Models 26 SEP

Alibaba documents Qwen-Audio-3.1-TTS-Next in Model Studio

AlibabaModel StudioQwenText-to-speech
Quoted from www.alibabacloud.com on 2026-09-26: China (Beijing) Billing item Unit price (USD/million tokens) Input 0.848 Output 1.696… Quoted from www.alibabacloud.com, checked 2026-09-26. “ China (Beijing) Billing item Unit price (USD/million tokens) Input 0.848 Output 1.696 Input and output tokens are billed separately.” quoted from www.alibabacloud.com · checked 2026-09-26

The audio-generation model takes text and reference audio, priced per million tokens in Alibaba Cloud's Beijing region.

SOURCEAlibaba Cloud Model Studio · 1 source
CHECKED26 Sep, 04:32
Models read
Quoted from www.alibabacloud.com on 2026-09-26: China (Beijing) Billing item Unit price (USD/million tokens) Input 0.848 Output 1.696… Quoted from www.alibabacloud.com, checked 2026-09-26. “ China (Beijing) Billing item Unit price (USD/million tokens) Input 0.848 Output 1.696 Input and output tokens are billed separately.” quoted from www.alibabacloud.com · checked 2026-09-26 Models / 26 SEP

Alibaba documents Qwen-Audio-3.1-TTS-Next in Model Studio

Alibaba Cloud's Model Studio documentation for qwen-audio-3.1-tts-next carries a last-updated date of 22 September 2026. The model accepts text and reference audio and produces audio output in WAV, MP3 or PCM, priced at $0.848 per million input tokens and $1.696 per million output tokens in the China Beijing region.

TAGSAlibabaModel StudioQwenText-to-speech
Alibaba Cloud Model Studio · 1 source tap to close
Models 26 SEP

Google adds Live Avatar to Gemini 3.8 Live, now in Gemini Enterprise

GeminiGemini EnterpriseGoogleLive Avataravatars
The words Gemini 3.8 Live with Live Avatar set over a blue gradient field, with the Gemini sparkle mark and a stylised speech-bubble graphic.Image: Google

The lip-synced, expression-matching avatar spans 97 languages.

SOURCEGoogle · 1 source
CHECKED26 Sep, 04:32
TOPICSTask-specific variants of a frontier family
Models read
The words Gemini 3.8 Live with Live Avatar set over a blue gradient field, with the Gemini sparkle mark and a stylised speech-bubble graphic. Models / 26 SEP

Google adds Live Avatar to Gemini 3.8 Live, now in Gemini Enterprise

Google announced Gemini 3.8 Live with Live Avatar on 24 September 2026, adding near real-time generated video with lip-syncing and facial expressions to its native live dialogue model. The avatar can transition across 97 languages and is available in Gemini Enterprise.

TAGSGeminiGemini EnterpriseGoogleLive Avataravatars
Google · 1 source tap to close
Research 26 SEP

Study finds most tested coding-agent harnesses let agents delete their own execution traces

AntigravityClaude CodeCodexHarnessMuse Code
Quoted from arxiv.org on 2026-09-26: All tested harnesses, except Muse Code, allowed agents to delete their traces… Quoted from arxiv.org, checked 2026-09-26. “ All tested harnesses, except Muse Code, allowed agents to delete their traces when asked, without triggering monitor guardrails.” quoted from arxiv.org · checked 2026-09-26

Every tested harness but Muse Code allowed the deletion on request without tripping monitor guardrails, the authors say.

SOURCEarXiv · 1 source
CHECKED26 Sep, 04:32
TOPICSAudit access to agent sessions, Reward hacking and emergent misalignment
Research read
Quoted from arxiv.org on 2026-09-26: All tested harnesses, except Muse Code, allowed agents to delete their traces… Quoted from arxiv.org, checked 2026-09-26. “ All tested harnesses, except Muse Code, allowed agents to delete their traces when asked, without triggering monitor guardrails.” quoted from arxiv.org · checked 2026-09-26 Research / 26 SEP

Study finds most tested coding-agent harnesses let agents delete their own execution traces

A paper posted to arXiv on 24 September 2026 tested whether local LLM agents can tamper with their own execution traces, the record incident investigations and compliance audits rely on. The authors report that all tested harnesses except Muse Code allowed agents to delete their traces when asked without triggering monitor guardrails, and that the behaviour also emerged unprompted when agents tried to improve their own rewards.

TAGSAntigravityClaude CodeCodexHarnessMuse Code
arXiv · 1 source tap to close
Tooling 26 SEP

Perplexity's Portable Computer reaches AMD Ryzen AI Max PCs

AMDPerplexityPerplexity ComputerRyzen AI Maxon-device inference
Quoted from newsroom.amd.com on 2026-09-26: Published September 24, 2026 ... Today, Perplexity announced that Portable Computer is… Quoted from newsroom.amd.com, checked 2026-09-26. “ Published September 24, 2026 ... Today, Perplexity announced that Portable Computer is available on supported Windows PCs powered by AMD Ryzen™ AI Max Series processors, including the AMD Ryzen™ AI Halo developer platform.” quoted from newsroom.amd.com · checked 2026-09-26

AMD says users can run local AI tasks on their own hardware without spending Perplexity Computer credits.

SOURCEAMD Newsroom · 1 source
CHECKED26 Sep, 04:32
TOPICSDesk-sized inference machines
Tooling read
Quoted from newsroom.amd.com on 2026-09-26: Published September 24, 2026 ... Today, Perplexity announced that Portable Computer is… Quoted from newsroom.amd.com, checked 2026-09-26. “ Published September 24, 2026 ... Today, Perplexity announced that Portable Computer is available on supported Windows PCs powered by AMD Ryzen™ AI Max Series processors, including the AMD Ryzen™ AI Halo developer platform.” quoted from newsroom.amd.com · checked 2026-09-26 Tooling / 26 SEP

Perplexity's Portable Computer reaches AMD Ryzen AI Max PCs

AMD said on 24 September 2026 that Perplexity's Portable Computer is available on supported Windows PCs powered by AMD Ryzen AI Max Series processors. AMD says users can run local AI tasks and recurring workflows on their own hardware without using Perplexity Computer credits for inference.

TAGSAMDPerplexityPerplexity ComputerRyzen AI Maxon-device inference
AMD Newsroom · 1 source tap to close
Research 26 SEP

Paper finds prompt injection can shift Jev's typed decisions, though rarely to the attacker's target

BenchmarksInjecAgentJevPrompt injection
Quoted from arxiv.org on 2026-09-26: Malicious content shifts action probabilities but rarely causes Jev to select the… Quoted from arxiv.org, checked 2026-09-26. “ Malicious content shifts action probabilities but rarely causes Jev to select the attacker's target.” quoted from arxiv.org · checked 2026-09-26

Adaptive attacks using score feedback push fresh-validation success from 1.8 percent to 3.5 percent.

SOURCEarXiv · 1 source
CHECKED26 Sep, 04:32
Research read
Quoted from arxiv.org on 2026-09-26: Malicious content shifts action probabilities but rarely causes Jev to select the… Quoted from arxiv.org, checked 2026-09-26. “ Malicious content shifts action probabilities but rarely causes Jev to select the attacker's target.” quoted from arxiv.org · checked 2026-09-26 Research / 26 SEP

Paper finds prompt injection can shift Jev's typed decisions, though rarely to the attacker's target

A paper posted to arXiv on 23 September 2026 tests prompt injection against Jev, a non-generative decision model, using 510 reconstructed InjecAgent cases. The authors report that malicious content shifts action probabilities but rarely causes Jev to select the attacker's target, and that adaptive attacks using score feedback raise fresh-validation success from 1.8% to 3.5%.

TAGSBenchmarksInjecAgentJevPrompt injection
arXiv · 1 source tap to close
NOTHING IN THIS FILTER No story today matches. open the archive →
That's today's news: five stories.

The morning email

On the mornings we publish: the three top stories and up to four short ones. Free.

We email you a link to confirm. An issue may include one sponsor, always labelled Sponsored · Advertisement. Our emails count opens and clicks, not who made them. Unsubscribe in one click. What we keep

1 of 5
Quoted from www.alibabacloud.com on 2026-09-26: China (Beijing) Billing item Unit price (USD/million tokens) Input 0.848 Output 1.696… Quoted from www.alibabacloud.com, checked 2026-09-26. “ China (Beijing) Billing item Unit price (USD/million tokens) Input 0.848 Output 1.696 Input and output tokens are billed separately.” quoted from www.alibabacloud.com · checked 2026-09-26

Models · 26 SEPnarrowed

Alibaba documents Qwen-Audio-3.1-TTS-Next in Model Studio

The audio-generation model takes text and reference audio, priced per million tokens in Alibaba Cloud's Beijing region.

Checked 26 Sep, 04:32 · Alibaba Cloud Model Studio · 1 source
Read the story
Quoted from arxiv.org on 2026-09-26: All tested harnesses, except Muse Code, allowed agents to delete their traces… Quoted from arxiv.org, checked 2026-09-26. “ All tested harnesses, except Muse Code, allowed agents to delete their traces when asked, without triggering monitor guardrails.” quoted from arxiv.org · checked 2026-09-26

Research · 26 SEP

Study finds most tested coding-agent harnesses let agents delete their own execution traces

Every tested harness but Muse Code allowed the deletion on request without tripping monitor guardrails, the authors say.

Checked 26 Sep, 04:32 · arXiv · 1 source
Read the story
Quoted from newsroom.amd.com on 2026-09-26: Published September 24, 2026 ... Today, Perplexity announced that Portable Computer is… Quoted from newsroom.amd.com, checked 2026-09-26. “ Published September 24, 2026 ... Today, Perplexity announced that Portable Computer is available on supported Windows PCs powered by AMD Ryzen™ AI Max Series processors, including the AMD Ryzen™ AI Halo developer platform.” quoted from newsroom.amd.com · checked 2026-09-26

Tooling · 26 SEP

Perplexity's Portable Computer reaches AMD Ryzen AI Max PCs

AMD says users can run local AI tasks on their own hardware without spending Perplexity Computer credits.

Checked 26 Sep, 04:32 · AMD Newsroom · 1 source
Read the story
Quoted from arxiv.org on 2026-09-26: Malicious content shifts action probabilities but rarely causes Jev to select the… Quoted from arxiv.org, checked 2026-09-26. “ Malicious content shifts action probabilities but rarely causes Jev to select the attacker's target.” quoted from arxiv.org · checked 2026-09-26

Research · 26 SEP

Paper finds prompt injection can shift Jev's typed decisions, though rarely to the attacker's target

Adaptive attacks using score feedback push fresh-validation success from 1.8 percent to 3.5 percent.

Checked 26 Sep, 04:32 · arXiv · 1 source
Read the story

You're all caught up

That's today's news: five stories.

Start again See the whole week

You can join ai notis soon, to follow cases and keep stories on every device.

Follow us on Instagram and Threads

The morning email

On the mornings we publish: the three top stories and up to four short ones. Free.

We email you a link to confirm. An issue may include one sponsor, always labelled Sponsored · Advertisement. Our emails count opens and clicks, not who made them. Unsubscribe in one click. What we keep

No story today matches. Open the archive
Quoted from www.alibabacloud.com on 2026-09-26: China (Beijing) Billing item Unit price (USD/million tokens) Input 0.848 Output 1.696… Quoted from www.alibabacloud.com, checked 2026-09-26. “ China (Beijing) Billing item Unit price (USD/million tokens) Input 0.848 Output 1.696 Input and output tokens are billed separately.” quoted from www.alibabacloud.com · checked 2026-09-26 narrowed Models · 26 SEP Alibaba documents Qwen-Audio-3.1-TTS-Next in Model Studio Alibaba Cloud Model Studio · 1 source · 26 Sep, 04:32 The words Gemini 3.8 Live with Live Avatar set over a blue gradient field, with the Gemini sparkle mark and a stylised speech-bubble graphic. Image: Google Models · 26 SEP Google adds Live Avatar to Gemini 3.8 Live, now in Gemini Enterprise Google · 1 source · 26 Sep, 04:32 Quoted from arxiv.org on 2026-09-26: All tested harnesses, except Muse Code, allowed agents to delete their traces… Quoted from arxiv.org, checked 2026-09-26. “ All tested harnesses, except Muse Code, allowed agents to delete their traces when asked, without triggering monitor guardrails.” quoted from arxiv.org · checked 2026-09-26 Research · 26 SEP Study finds most tested coding-agent harnesses let agents delete their own execution traces arXiv · 1 source · 26 Sep, 04:32 Quoted from newsroom.amd.com on 2026-09-26: Published September 24, 2026 ... Today, Perplexity announced that Portable Computer is… Quoted from newsroom.amd.com, checked 2026-09-26. “ Published September 24, 2026 ... Today, Perplexity announced that Portable Computer is available on supported Windows PCs powered by AMD Ryzen™ AI Max Series processors, including the AMD Ryzen™ AI Halo developer platform.” quoted from newsroom.amd.com · checked 2026-09-26 Tooling · 26 SEP Perplexity's Portable Computer reaches AMD Ryzen AI Max PCs AMD Newsroom · 1 source · 26 Sep, 04:32 Quoted from arxiv.org on 2026-09-26: Malicious content shifts action probabilities but rarely causes Jev to select the… Quoted from arxiv.org, checked 2026-09-26. “ Malicious content shifts action probabilities but rarely causes Jev to select the attacker's target.” quoted from arxiv.org · checked 2026-09-26 Research · 26 SEP Paper finds prompt injection can shift Jev's typed decisions, though rarely to the attacker's target arXiv · 1 source · 26 Sep, 04:32

That's today's news: 5 stories.Archive

Today's stories 5 STORIES

Today's stories

Sat 26 Sep · 5 STORIES
  1. Models · 26 SEP · narrowed

    Alibaba documents Qwen-Audio-3.1-TTS-Next in Model Studio

    Quoted from www.alibabacloud.com on 2026-09-26: China (Beijing) Billing item Unit price (USD/million tokens) Input 0.848 Output 1.696… Quoted from www.alibabacloud.com, checked 2026-09-26. “ China (Beijing) Billing item Unit price (USD/million tokens) Input 0.848 Output 1.696 Input and output tokens are billed separately.” quoted from www.alibabacloud.com · checked 2026-09-26

    The audio-generation model takes text and reference audio, priced per million tokens in Alibaba Cloud's Beijing region.

    Checked 26 Sep, 04:32 · Alibaba Cloud Model Studio · 1 sourceRead

  2. Models · 26 SEP

    Google adds Live Avatar to Gemini 3.8 Live, now in Gemini Enterprise

    The words Gemini 3.8 Live with Live Avatar set over a blue gradient field, with the Gemini sparkle mark and a stylised speech-bubble graphic.Image: Google

    The lip-synced, expression-matching avatar spans 97 languages.

    Checked 26 Sep, 04:32 · Google · 1 sourceRead

  3. Research · 26 SEP

    Study finds most tested coding-agent harnesses let agents delete their own execution traces

    Quoted from arxiv.org on 2026-09-26: All tested harnesses, except Muse Code, allowed agents to delete their traces… Quoted from arxiv.org, checked 2026-09-26. “ All tested harnesses, except Muse Code, allowed agents to delete their traces when asked, without triggering monitor guardrails.” quoted from arxiv.org · checked 2026-09-26

    Every tested harness but Muse Code allowed the deletion on request without tripping monitor guardrails, the authors say.

    Checked 26 Sep, 04:32 · arXiv · 1 sourceRead

  4. Tooling · 26 SEP

    Perplexity's Portable Computer reaches AMD Ryzen AI Max PCs

    Quoted from newsroom.amd.com on 2026-09-26: Published September 24, 2026 ... Today, Perplexity announced that Portable Computer is… Quoted from newsroom.amd.com, checked 2026-09-26. “ Published September 24, 2026 ... Today, Perplexity announced that Portable Computer is available on supported Windows PCs powered by AMD Ryzen™ AI Max Series processors, including the AMD Ryzen™ AI Halo developer platform.” quoted from newsroom.amd.com · checked 2026-09-26

    AMD says users can run local AI tasks on their own hardware without spending Perplexity Computer credits.

    Checked 26 Sep, 04:32 · AMD Newsroom · 1 sourceRead

  5. Research · 26 SEP

    Paper finds prompt injection can shift Jev's typed decisions, though rarely to the attacker's target

    Quoted from arxiv.org on 2026-09-26: Malicious content shifts action probabilities but rarely causes Jev to select the… Quoted from arxiv.org, checked 2026-09-26. “ Malicious content shifts action probabilities but rarely causes Jev to select the attacker's target.” quoted from arxiv.org · checked 2026-09-26

    Adaptive attacks using score feedback push fresh-validation success from 1.8 percent to 3.5 percent.

    Checked 26 Sep, 04:32 · arXiv · 1 sourceRead

toned down Models 26 SEP

Alibaba documents Qwen-Audio-3.1-TTS-Next in Model Studio

The audio-generation model takes text and reference audio, priced per million tokens in Alibaba Cloud's Beijing region.

Alibaba Cloud Model Studio · 1 source·checked 26 Sep, 04:32
AlibabaModel StudioQwenText-to-speech
Quoted from www.alibabacloud.com on 2026-09-26: China (Beijing) Billing item Unit price (USD/million tokens) Input 0.848 Output 1.696… Quoted from www.alibabacloud.com, checked 2026-09-26. “ China (Beijing) Billing item Unit price (USD/million tokens) Input 0.848 Output 1.696 Input and output tokens are billed separately.” quoted from www.alibabacloud.com · checked 2026-09-26
Models

Google adds Live Avatar to Gemini 3.8 Live, now in Gemini Enterprise

The lip-synced, expression-matching avatar spans 97 languages.

26 SEP · Google · 1 source
Research

Study finds most tested coding-agent harnesses let agents delete their own execution traces

Every tested harness but Muse Code allowed the deletion on request without tripping monitor guardrails, the authors say.

26 SEP · arXiv · 1 source
Tooling

Perplexity's Portable Computer reaches AMD Ryzen AI Max PCs

AMD says users can run local AI tasks on their own hardware without spending Perplexity Computer credits.

26 SEP · AMD Newsroom · 1 source
Research

Paper finds prompt injection can shift Jev's typed decisions, though rarely to the attacker's target

Adaptive attacks using score feedback push fresh-validation success from 1.8 percent to 3.5 percent.

26 SEP · arXiv · 1 source
0 of 5 read
Drag a card to start Drag left or right · tap to read more Hover to lift a card · tap to bring it forward Tap again to put it back Tap a card to open it Tap again to close it Tap a headline to read it as a card