3signals21 Signals being tracked, weekly summary from the last 7 days:
Site: 3signals - X: @3signalsai
September 26, 2026
Share: X
This is the weekly summary of signals from the last 7 days. The 3 newest signals are first, followed by 18 more in reverse chronological order. Open the full signal list
Weekly summary: 3 new signals first
1. Fireworks Research launches Ember-1, a model matching Kimi K3's quality with 40% fewer tokens
model-releases - release - September 26, 2026
What changed? Fireworks - Blog Fireworks - Blog Join us for our inaugural conference, Forge 2026 Product Solutions Models Pricing Resources Log In Get Started Fireworks Blog Introducing Ember-1 Ember-1 is a new specialized model from Fireworks Research that delivers Kimi K3’s quality with 40% fewer tokens. [excerpt shortened].
Article: Fireworks Research launches Ember-1, a model matching Kimi K3's quality with 40% fewer tokens
From: fireworks-ai - source
Source context: Fireworks Research launches Ember-1, a model matching Kimi K3's quality with 40% fewer tokens. Evidence: Fireworks - Blog Fireworks - Blog Join us for our inaugural conference, Forge 2026 Product Solutions Models Pricing Resources Log In Get Started Fireworks Blog Introducing Ember-1 Ember-1 is a new specialized model from Fireworks Research that delivers Kimi K3’s quality with 40% fewer tokens. [excerpt shortened]
Excerpt: Fireworks - Blog Fireworks - Blog Join us for our inaugural conference, Forge 2026 Product Solutions Models Pricing Resources Log In Get Started Fireworks Blog Introducing Ember-1 Ember-1 is a new specialized model from Fireworks Research that delivers Kimi K3’s quality with 40% fewer tokens. [excerpt shortened]
Why is this signal important? This matters because model capability is shifting what builders can expect from current tools.
2. Datacor integrates Amazon Quick Sight for self-service rental analytics. (title shortened)
ai-products, inference-infrastructure - business, production - September 26, 2026
What changed? That approach had to make data broadly accessible while accommodating the diverse analytical needs of gas and welding distributors. “ “The interesting part wasn’t the AI—it was the unglamorous work underneath it: cross-cloud ingestion, row-count validation, a tenant-isolation model we could actually audit.
Article: Datacor integrates Amazon Quick Sight for self-service rental analytics. (title shortened)
From: aws - source
Source context: Datacor integrates Amazon Quick Sight for self-service rental analytics, enabling gas and welding distributors to access real-time insights without IT intervention. Evidence: That approach had to make data broadly accessible while accommodating the diverse analytical needs of gas and welding distributors. “ “The interesting part wasn’t the AI—it was the unglamorous work underneath it: cross-cloud ingestion, row-count validation, a tenant-isolation model we could actually audit.
Excerpt: That approach had to make data broadly accessible while accommodating the diverse analytical needs of gas and welding distributors. “ “The interesting part wasn’t the AI—it was the unglamorous work underneath it: cross-cloud ingestion, row-count validation, a tenant-isolation model we could actually audit.
Why is this signal important? This matters because voice AI is becoming more useful for live translation, transcription, and assistants.
3. Amazon SageMaker JumpStart enables real-time voice cloning with Qwen3-TTS-12Hz-1.
inference-infrastructure, ai-products - production, business - September 26, 2026
What changed? Deploying real-time personalized speech with Qwen3-TTS on Amazon SageMaker AI With voice cloning, you can generate new speech in a target speaker’s voice from a short reference recording, without retraining a model. You can now deploy the publicly available Qwen3-TTS-12Hz-1.7B-Base text-to-speech model from Amazon SageMaker JumpStart to a fully managed, real-time inference endpoint.
Article: Amazon SageMaker JumpStart enables real-time voice cloning with Qwen3-TTS-12Hz-1.
From: aws - source
Source context: Amazon SageMaker JumpStart enables real-time voice cloning with Qwen3-TTS-12Hz-1.7B-Base, supporting multilingual and personalized speech synthesis. Evidence: Deploying real-time personalized speech with Qwen3-TTS on Amazon SageMaker AI With voice cloning, you can generate new speech in a target speaker’s voice from a short reference recording, without retraining a model. You can now deploy the publicly available Qwen3-TTS-12Hz-1.7B-Base text-to-speech model from Amazon SageMaker JumpStart to a fully managed, real-time inference endpoint.
Excerpt: Deploying real-time personalized speech with Qwen3-TTS on Amazon SageMaker AI With voice cloning, you can generate new speech in a target speaker’s voice from a short reference recording, without retraining a model. You can now deploy the publicly available Qwen3-TTS-12Hz-1. [excerpt shortened]
Why is this signal important? This matters because voice AI is becoming more useful for live translation, transcription, and assistants.
4. NarrateAI implements five quality assurance techniques on Amazon Bedrock to ensure real-time. (title shortened)
ai-products, inference-infrastructure, agent-workflows - production, business, open-source - September 26, 2026
What changed? Closing it requires production-ready quality assurance built into every step, from data retrieval to response delivery. This post details five techniques, implemented on Amazon Bedrock , that work together to deliver it: adaptive pipeline orchestration, cross-account multi-model failover, real-time streaming evaluation, composite evaluation framework, and data accuracy verification.
From: aws - source
Source context: NarrateAI implements five quality assurance techniques on Amazon Bedrock to ensure real-time, accurate LLM responses for business intelligence. Evidence: Closing it requires production-ready quality assurance built into every step, from data retrieval to response delivery. This post details five techniques, implemented on Amazon Bedrock , that work together to deliver it: adaptive pipeline orchestration, cross-account multi-model failover, real-time streaming evaluation, composite evaluation framework, and data accuracy verification.
Excerpt: Closing it requires production-ready quality assurance built into every step, from data retrieval to response delivery. This post details five techniques, implemented on Amazon Bedrock , that work together to deliver it: adaptive pipeline orchestration, cross-account multi-model failover, real-time streaming evaluation, composite evaluation framework, and data accuracy verification.
Why is this signal important? This matters because frontier AI economics and compute needs are scaling quickly.
5. SkyRL accelerates multimodal RL training on Amazon SageMaker HyperPod. (title shortened)
agent-workflows, inference-infrastructure - release, production, open-source - September 26, 2026
What changed? SkyRL accelerates multimodal RL training on Amazon SageMaker HyperPod, improving maze solve rates from 43.75% to over 95%. Evidence: In this post, we show how to use these capabilities to run SkyRL , an open-source RL framework, to train a Qwen3-VL-8B vision-language model to navigate visual mazes using Group Relative Policy Optimization (GRPO) on SageMaker HyperPod. Starting from the VisGym SFT checkpoint , a supervised fine-tuning (SFT) starting point, GRPO post-training on HyperPod improves the maze solve rate from 43.75% to more than 95% on a fixed 64-maze evaluation set.
Article: SkyRL accelerates multimodal RL training on Amazon SageMaker HyperPod. (title shortened)
From: aws - source
Source context: SkyRL accelerates multimodal RL training on Amazon SageMaker HyperPod, improving maze solve rates from 43.75% to over 95%. Evidence: In this post, we show how to use these capabilities to run SkyRL , an open-source RL framework, to train a Qwen3-VL-8B vision-language model to navigate visual mazes using Group Relative Policy Optimization (GRPO) on SageMaker HyperPod. Starting from the VisGym SFT checkpoint , a supervised fine-tuning (SFT) starting point, GRPO post-training on HyperPod improves the maze solve rate from 43.75% to more than 95% on a fixed 64-maze evaluation set.
Excerpt: Starting from the VisGym SFT checkpoint , a supervised fine-tuning (SFT) starting point, GRPO post-training on HyperPod improves the maze solve rate from 43.75% to more than 95% on a fixed 64-maze evaluation set. [excerpt shortened]
Why is this signal important? This matters because frontier AI economics and compute needs are scaling quickly.
6. LangSmith Custom Apps enables custom interface creation for agent data without hosting hassles
agent-workflows - production, open-source, release - September 26, 2026
What changed? LangSmith Custom Apps: Build custom interfaces around your agent data LangSmith Custom Apps lets you build the interface you want with your LangSmith data, publish it into your workspace, and skip the hosting, auth, and permissions work. Learn more.
Article: LangSmith Custom Apps enables custom interface creation for agent data without hosting hassles
From: langchain - source
Source context: LangSmith Custom Apps enables custom interface creation for agent data without hosting hassles. Evidence: LangSmith Custom Apps: Build custom interfaces around your agent data LangSmith Custom Apps lets you build the interface you want with your LangSmith data, publish it into your workspace, and skip the hosting, auth, and permissions work. Learn more.
Excerpt: LangSmith Custom Apps: Build custom interfaces around your agent data LangSmith Custom Apps lets you build the interface you want with your LangSmith data, publish it into your workspace, and skip the hosting, auth, and permissions work. Learn more.
Why is this signal important? This matters because open-source AI tooling is becoming a larger part of production engineering work.
7. Vironexis reports promising results with VNX-101, a liver-based cancer drug factory
ai-products - research, business - September 26, 2026
What changed? This week, a company called Vironexis shared encouraging results from putting those two ideas together: give the body instructions to make its own cancer-fighting medicine. On Monday, Vironexis reported that all three patients in a subgroup with relapsed or refractory acute lymphoblastic leukemia achieved complete responses after a single administration of VNX-101, with no residual leukemia detected in testing.
Article: Vironexis reports promising results with VNX-101, a liver-based cancer drug factory
From: packy-mccormick - source
Source context: Vironexis reports promising results with VNX-101, a liver-based cancer drug factory. Evidence: This week, a company called Vironexis shared encouraging results from putting those two ideas together: give the body instructions to make its own cancer-fighting medicine. On Monday, Vironexis reported that all three patients in a subgroup with relapsed or refractory acute lymphoblastic leukemia achieved complete responses after a single administration of VNX-101, with no residual leukemia detected in testing.
Excerpt: This week, a company called Vironexis shared encouraging results from putting those two ideas together: give the body instructions to make its own cancer-fighting medicine. On Monday, Vironexis reported that all three patients in a subgroup with relapsed or refractory acute lymphoblastic leukemia achieved complete responses after a single administration. [excerpt shortened]
Why is this signal important? This matters because Vironexis reports promising results with VNX-101, a liver-based cancer drug factory.
8. Meta's Muse app quickly tops US charts but raises questions about user loyalty and economic alignment
agent-workflows, ai-products - business, production, open-source, release - September 26, 2026
What changed? But is Muse my butler, or is it secretly working for Meta? That small fee, from merchant to Meta, is a big problem.
From: azeem-azhar - source
Source context: Meta's Muse app quickly tops US charts but raises questions about user loyalty and economic alignment. Evidence: But is Muse my butler, or is it secretly working for Meta? That small fee, from merchant to Meta, is a big problem.
Excerpt: But is Muse my butler, or is it secretly working for Meta? That small fee, from merchant to Meta, is a big problem.
Why is this signal important? This matters because open-source AI tooling is becoming a larger part of production engineering work.
9. OpenRouter becomes critical AI infrastructure by enabling multi-model routing for over 10 million. (title shortened)
ai-products, inference-infrastructure - business, production - September 26, 2026
What changed? Alex Atallah [00:06:35]: Yeah. Anjney Midha [00:06:35]: Yeah.
From: alessio-fanelli - source
Source context: OpenRouter becomes critical AI infrastructure by enabling multi-model routing for over 10 million developers, highlighting the importance of model diversity and distribution. Evidence: Alex Atallah [00:06:35]: Yeah. Anjney Midha [00:06:35]: Yeah.
Excerpt: Alex Atallah [00:06:35]: Yeah. Anjney Midha [00:06:35]: Yeah.
Why is this signal important? This matters because serving improvements can make AI products faster and cheaper to run.
10. Jensen Huang calls for increased AI safety spending and suggests shutting down labs if AI models can't be contained
ai-safety - safety, research - September 26, 2026
What changed? Jensen Huang: Now, if they say the alternative, which is: There is no way to contain our experiments, there’s just no way; when we test our A.I. models, it will get out, and it will damage the world — then I think the answer is that we have to shut the labs down.
From: zvi-mowshowitz - source
Source context: Jensen Huang calls for increased AI safety spending and suggests shutting down labs if AI models can't be contained. Evidence: Jensen Huang: Now, if they say the alternative, which is: There is no way to contain our experiments, there’s just no way; when we test our A.I. models, it will get out, and it will damage the world — then I think the answer is that we have to shut the labs down.
Excerpt: models, it will get out, and it will damage the world — then I think the answer is that we have to shut the labs down. Because the cost to humanity, the damage is too great.
Why is this signal important? This matters because model capability is shifting what builders can expect from current tools.
11. Meta's Muse offers a consumer-accessible AI with persistent Linux VMs, raising safety concerns
ai-products, agent-workflows, ai-safety, model-releases - business, release, open-source, production - September 26, 2026
What changed? [...] I don’t think people realize how powerful — and thus dangerous — Muse is, especially if it’s running on your Mac. — John Gruber , Muse Looks Cute, but Looks are Deceiving Tags: meta , ai , llms , general-agents , generative-ai , john-gruber , muse-agent , muse.
Article: Meta's Muse offers a consumer-accessible AI with persistent Linux VMs, raising safety concerns
From: simon-willison - source
Source context: Meta's Muse offers a consumer-accessible AI with persistent Linux VMs, raising safety concerns. Evidence: [...] I don’t think people realize how powerful — and thus dangerous — Muse is, especially if it’s running on your Mac. — John Gruber , Muse Looks Cute, but Looks are Deceiving Tags: meta , ai , llms , general-agents , generative-ai , john-gruber , muse-agent , muse
Excerpt: [...] I don’t think people realize how powerful — and thus dangerous — Muse is, especially if it’s running on your Mac. — John Gruber , Muse Looks Cute, but Looks are Deceiving Tags: meta , ai , llms , general-agents , generative-ai , john-gruber , muse-agent , muse
Why is this signal important? This matters because Meta's Muse offers a consumer-accessible AI with persistent Linux VMs, raising safety concerns.
12. Managed Deep Agents 0.8 enhances production agent deployment with user-owned credentials and new tools
agent-workflows - release, production, open-source - September 25, 2026
What changed? The 0.8 release adds support for user-owned credentials, user-level memory, HTTP channels, file transfer in Slack and a pre-built tool for web search.
From: langchain - source
Source context: Managed Deep Agents 0.8 enhances production agent deployment with user-owned credentials and new tools. Evidence: The 0.8 release adds support for user-owned credentials, user-level memory, HTTP channels, file transfer in Slack and a pre-built tool for web search.
Excerpt: The 0.8 release adds support for user-owned credentials, user-level memory, HTTP channels, file transfer in Slack and a pre-built tool for web search.
Why is this signal important? This matters because open-source AI tooling is becoming a larger part of production engineering work.
13. LangChain updates LangSmith with Engine v2, Managed Deep Agents, and fine-tuning capabilities
agent-workflows, ai-safety - release, safety, production, open-source - September 25, 2026
What changed? New in LangSmith: Engine v2, Managed Deep Agents, Fine-Tuning, and more LangChain announced new updates to LangSmith. Updates include Engine v2 with red teaming and automatic testing, a new version of Managed Deep Agents, trajectories and more.
Article: LangChain updates LangSmith with Engine v2, Managed Deep Agents, and fine-tuning capabilities
From: langchain - source
Source context: LangChain updates LangSmith with Engine v2, Managed Deep Agents, and fine-tuning capabilities. Evidence: New in LangSmith: Engine v2, Managed Deep Agents, Fine-Tuning, and more LangChain announced new updates to LangSmith. Updates include Engine v2 with red teaming and automatic testing, a new version of Managed Deep Agents, trajectories and more.
Excerpt: New in LangSmith: Engine v2, Managed Deep Agents, Fine-Tuning, and more LangChain announced new updates to LangSmith. Updates include Engine v2 with red teaming and automatic testing, a new version of Managed Deep Agents, trajectories and more.
Why is this signal important? This matters because open-source AI tooling is becoming a larger part of production engineering work.
14. Gemini 3.8 Live launches with a new Live Avatar feature
ai-products - release, business - September 25, 2026
What changed? Gemini 3.8 Live now includes a Live Avatar feature, enhancing user interaction. The signal is supported by 2 sources, including google-deepmind.
Article: Gemini 3.8 Live launches with a new Live Avatar feature
From: google-deepmind - source
Source context: Gemini 3.8 Live launches with a new Live Avatar feature.
Excerpt: Gemini 3.8 Live now includes a Live Avatar feature, enhancing user interaction.
Article: Gemini 3.8 introduces text-to-speech capabilities
From: google-deepmind - source
Source context: Evidence: Gemini 3.8 now includes text-to-speech functionality, enhancing its communication features.
Excerpt: Gemini 3.8 now includes text-to-speech functionality, enhancing its communication features.
Why is this signal important? This matters because Gemini 3.8 Live launches with a new Live Avatar feature.
15. Muse's consumer AI agent gains traction with personalized, approachable design and practical utility
ai-products - business, release - September 25, 2026
What changed? Everyone knows how to text already! Muse adds a few more touches to increase its approachability and build upon most folks’ existing mental model of how a helpful human works.
Article: Muse's consumer AI agent gains traction with personalized, approachable design and practical utility
From: julie-zhuo - source
Source context: Muse's consumer AI agent gains traction with personalized, approachable design and practical utility. Evidence: Everyone knows how to text already! Muse adds a few more touches to increase its approachability and build upon most folks’ existing mental model of how a helpful human works.
Excerpt: Everyone knows how to text already! Muse adds a few more touches to increase its approachability and build upon most folks’ existing mental model of how a helpful human works.
Why is this signal important? This matters because teams are turning AI agents into repeatable production workflows.
16. AI transforms scientific research by enabling faster data analysis and decision-making. (title shortened)
ai-products, ai-safety, model-releases - business, safety, research, production - September 25, 2026
What changed? Experiments iterate faster when analysis takes an hour instead of a week. A category of software that would have been licensed for six figures becomes a one-day build.
From: alessio-fanelli - source
Source context: AI transforms scientific research by enabling faster data analysis and decision-making, exemplified by Endura Therapeutics' innovative approaches. Evidence: Experiments iterate faster when analysis takes an hour instead of a week. A category of software that would have been licensed for six figures becomes a one-day build.
Excerpt: Experiments iterate faster when analysis takes an hour instead of a week. A category of software that would have been licensed for six figures becomes a one-day build.
Why is this signal important? This matters because AI transforms scientific research by enabling faster data analysis and decision-making, exemplified (shortened).
17. Anthropic's Claude Opus 5.5 matches Claude Fable 5.1's performance while reducing costs by 40%
model-releases, ai-safety, evaluations, ai-products - release, safety, research, production - September 24, 2026
What changed? Newsroom \ Anthropic Newsroom \ Anthropic Skip to main content Skip to footer Research Policy Commitments Learn News Try Claude Newsroom Press inquiries press@anthropic.com Non-media inquiries How to get support Media assets Download press kit Introducing Claude Opus 5.5 Announcements Sep 22, 2026 Opus 5.5 performs at the level of Claude Fable 5.1 on most work and costs 40% less to run than Opus 5. [excerpt shortened] The signal is supported by 2 sources, including anthropic, alessio-fanelli.
Article: Anthropic's Claude Opus 5.5 matches Claude Fable 5.1's performance while reducing costs by 40%
From: anthropic - source
Source context: Anthropic's Claude Opus 5.5 matches Claude Fable 5.1's performance while reducing costs by 40%. Evidence: Newsroom \ Anthropic Newsroom \ Anthropic Skip to main content Skip to footer Research Policy Commitments Learn News Try Claude Newsroom Press inquiries press@anthropic.com Non-media inquiries How to get support Media assets Download press kit Introducing Claude Opus 5.5 Announcements Sep 22, 2026 Opus 5.5 performs at the level of Claude Fable 5.1 on most work and costs 40% less to run than Opus 5. [excerpt shortened]
Excerpt: Newsroom \ Anthropic Newsroom \ Anthropic Skip to main content Skip to footer Research Policy Commitments Learn News Try Claude Newsroom Press inquiries press@anthropic.com Non-media inquiries How to get support Media assets Download press kit Introducing Claude Opus 5.5 Announcements Sep 22, 2026 Opus 5. [excerpt shortened]
From: alessio-fanelli - source
Source context: Anthropic launches Claude Opus 5.5, offering 40% cost reduction and improved writing over previous models. Evidence: Launch claims. Opus 5.5 “performs at the level of Claude Fable 5.1 for most tasks, and costs 40% less to run than Opus 5” ( @claudeai ; @AnthropicAI ).
Excerpt: Launch claims. Opus 5.5 “performs at the level of Claude Fable 5.1 for most tasks, and costs 40% less to run than Opus 5” ( @claudeai ; @AnthropicAI ).
Why is this signal important? This matters because model capability is shifting what builders can expect from current tools.
18. Anthropic's Claude Opus 5.5 claims superior performance over Fable 5.1 while being more cost-effective than Opus 5
ai-safety, ai-products, model-releases, evaluations - safety, research, business, release - September 24, 2026
What changed? Anthropic is claiming Opus 5.5 is outright as good or better than Fable 5.1, while being actively cheaper than Opus 5. That means it’s time for a good old system card reading.
From: zvi-mowshowitz - source
Source context: Anthropic's Claude Opus 5.5 claims superior performance over Fable 5.1 while being more cost-effective than Opus 5. Evidence: Anthropic is claiming Opus 5.5 is outright as good or better than Fable 5.1, while being actively cheaper than Opus 5. That means it’s time for a good old system card reading.
Excerpt: Anthropic is claiming Opus 5.5 is outright as good or better than Fable 5.1, while being actively cheaper than Opus 5. That means it’s time for a good old system card reading.
Why is this signal important? This matters because frontier AI economics and compute needs are scaling quickly.
19. Google launches Gemini 3.8 TTS models with 2,000 voices and custom voice creation
model-releases - release - September 24, 2026
What changed? Gemini 3.8 TTS Playground Tool: Gemini 3.8 TTS Playground Google released two new Gemini text-to-speech models today - gemini-3.8-flash-tts and gemini-3.8-flash-lite-tts . They come with a library of over 2,000 voices, plus the ability to create a custom voice with "just a 30-second audio sample of your voice or a voice you have the rights to use".
Article: Google launches Gemini 3.8 TTS models with 2,000 voices and custom voice creation
From: simon-willison - source
Source context: Google launches Gemini 3.8 TTS models with 2,000 voices and custom voice creation. Evidence: Gemini 3.8 TTS Playground Tool: Gemini 3.8 TTS Playground Google released two new Gemini text-to-speech models today - gemini-3.8-flash-tts and gemini-3.8-flash-lite-tts . They come with a library of over 2,000 voices, plus the ability to create a custom voice with "just a 30-second audio sample of your voice or a voice you have the rights to use".
Excerpt: Gemini 3.8 TTS Playground Tool: Gemini 3.8 TTS Playground Google released two new Gemini text-to-speech models today - gemini-3.8-flash-tts and gemini-3.8-flash-lite-tts . [excerpt shortened]
Why is this signal important? This matters because voice AI is becoming more useful for live translation, transcription, and assistants.
20. Jev integrates with LangSmith Evals for efficient agent trace evaluation
evaluations - production, research - September 23, 2026
What changed? Jev is now available in LangSmith Evals Use Jev as a judge for LangSmith evals to evaluate agent traces with faster, cheaper structured feedback across production runs, datasets, and regression tests. The signal is supported by 2 sources, including langchain.
Article: Jev integrates with LangSmith Evals for efficient agent trace evaluation
From: langchain - source
Source context: Jev integrates with LangSmith Evals for efficient agent trace evaluation. Evidence: Jev is now available in LangSmith Evals Use Jev as a judge for LangSmith evals to evaluate agent traces with faster, cheaper structured feedback across production runs, datasets, and regression tests.
Excerpt: Jev is now available in LangSmith Evals Use Jev as a judge for LangSmith evals to evaluate agent traces with faster, cheaper structured feedback across production runs, datasets, and regression tests.
Article: Jev-as-a-Judge outperforms LLM judges in agent evaluation metrics
From: langchain - source
Source context: Jev-as-a-Judge outperforms LLM judges in agent evaluation metrics. Evidence: We tested using Jev-as-a-Judge against LLM judges on accuracy, repeatability, latency, and cost to see whether System One models could offer a new approach to agent evaluation.
Excerpt: We tested using Jev-as-a-Judge against LLM judges on accuracy, repeatability, latency, and cost to see whether System One models could offer a new approach to agent evaluation.
Why is this signal important? This matters because evaluation practices are becoming more concrete for teams shipping AI agents.
21. OpenAI's GPT-6 Sol and Luna models launch at half the price of their predecessors, intensifying the AI model price war
model-releases, inference-infrastructure - release, production, business, open-source - September 23, 2026
What changed? Grok 4.7 priced itself at $2/$6, less than half the price of GPT-5.6 Sol, but is now equally priced to GPT-6 Sol on input and closer on output. At $0.10/$0.50 GPT-6 Luna is one of the cheapest models OpenAI have ever released, beaten only by the far weaker GPT-4.1 Nano ($0.10/$0.40, April 2025) and GPT-5 Nano ($0.05/$0.40, August 2025).
From: simon-willison - source
Source context: OpenAI's GPT-6 Sol and Luna models launch at half the price of their predecessors, intensifying the AI model price war. Evidence: Grok 4.7 priced itself at $2/$6, less than half the price of GPT-5.6 Sol, but is now equally priced to GPT-6 Sol on input and closer on output. At $0.10/$0.50 GPT-6 Luna is one of the cheapest models OpenAI have ever released, beaten only by the far weaker GPT-4.1 Nano ($0.10/$0.40, April 2025) and GPT-5 Nano ($0.05/$0.40, August 2025).
Excerpt: Grok 4.7 priced itself at $2/$6, less than half the price of GPT-5.6 Sol, but is now equally priced to GPT-6 Sol on input and closer on output. At $0.10/$0.50 GPT-6 Luna is one of the cheapest models OpenAI have ever released, beaten only by the far weaker GPT-4. [excerpt shortened]
Why is this signal important? This matters because model capability is shifting what builders can expect from current tools.
What's new with 3signals
Recent product improvements:
- Vibe Check section (2026-06-11): 3signals now has a Vibe Check section for surfacing community-validated momentum alongside the system's curated signal picks. Details
- Interactive wiki graph view (2026-05-18): The 3signals wiki now includes an Obsidian-style graph for exploring how signals connect to topics, concepts, authors, and source evidence. Details
- Front-end and back-end split for faster site delivery (2026-05-17): 3signals now serves the public website from Vercel while Railway keeps running the API, cron jobs, and content generation pipeline. Details
Staged future improvements:
- Fold reader feedback into presentation scoring so useful signals can be resurfaced with better timing.
- Expand archive analytics so opens, votes, site access, and X posts can be compared by issue.
- Continue tightening source QA for headline strength, evidence fit, and source freshness.