22 releases covered live on the show, led by GPT-6 Astra, Claude Fable 5.1 & Mythos 5.1 — every model, product, paper and tool that mattered, with links and our analysis.
22 AI releases shipped in September 2026, 22 of them in the latest week (Sep 3, 2026), led by GPT-6 Astra, Claude Fable 5.1 & Mythos 5.1, Muse Spark 1.3, Qwen3.8-Max-0902, MiniMax H3 Max Turbo. ThursdAI — the weekly AI news podcast hosted by Alex Volkov — tracked every entry below with source links, key numbers and episode coverage (all covered live on the show).
OpenAI launches GPT-6 Astra, calls it its most intelligent and aligned model. OpenAI released GPT-6 Astra on September 3, 2026, live during the ThursdAI stream, with president Greg Brockman telling reporters 'welcome to the AGI era.' OpenAI reports 99.9 on ARC-AGI-3, 97.6% on FrontierMath Tier 4, 72.6% on OSWorld 2.0, 92.7 on ScreenSpot Pro and 64.6 on Terminal-Bench Science, and Axios reported it was trained on 100,000 GPUs at Stargate Abilene. ThursdAI covered it live on the show, with primary sources linked below.
1 open-weights model shipped in September 2026, led by Hy4 preview (770B / 49B total / active parameters). Each entry below links the weights and the episode segment where we covered it.
14 companies shipped AI releases in September 2026; the most active were Meta AI, OpenAI, Anthropic, fal, Google DeepMind, Runway. Every launch below has primary-source links and ThursdAI's live episode analysis.
| Release | Best for | Why it matters | Key number |
|---|---|---|---|
| GPT-6 Astra · OpenAI | Agent builders | OpenAI launches GPT-6 Astra, calls it its most intelligent and aligned model | 99.9 ARC-AGI-3 (Sol was 7%) |
| Claude Fable 5.1 & Mythos 5.1 · Anthropic | Developers & coding agents | Claude Fable 5.1 & Mythos 5.1: Terminal-Bench 4.0 jumps to 55.8, cache reads 75% cheaper | 55.8 Terminal-Bench 4.0 (Fable 5: 42.0) |
| Muse Spark 1.3 · Meta AI | Developers & coding agents | Meta Muse Spark 1.3 ties GPT-5.6 Sol and Grok 4.6 on the AA index, max mode ties Fable 5 | 61 / 62 AA Intelligence Index, xhigh / max |
| Qwen3.8-Max-0902 · Alibaba Qwen | Developers & coding agents | Qwen3.8-Max-0902: 2.4T API-only refresh claims #1 on Code Arena | 2.4T parameters |
| MiniMax H3 Max Turbo · fal | Video creators | fal H3 Max Turbo: ~97% of H3 Max quality in 1.4 seconds at $0.01 per second | $0.01/sec price at 768p |
| Gemini 3.8 Flash · Google DeepMind | Developers & coding agents | Gemini 3.8 Flash: third Flash in three weeks, HLE-Verified 54.9, 1M context | 54.9 HLE-Verified |
| Realtime TTS-2 · Inworld | Voice-agent builders | Inworld Realtime TTS-2 goes GA: sub-100ms, $25/M characters, #1 on AA Controlled Voice Arena | <100ms latency |
| MAI-Transcribe-2 · Microsoft AI | Voice-agent builders | Microsoft MAI-Transcribe-2: 2.0% WER, 400x real time, $1.67 per 1,000 minutes | 2.0% WER, #2 on Artificial Analysis |
| Hy4 preview · Tencent | Open-source builders | Tencent Hy4 preview: 770B/49B Apache 2.0 MoE, Sherry quant shrinks 1.5 TB to 214 GB | 770B / 49B total / active parameters |
| Atlas · World Labs | Video creators | World Labs Atlas: omnimodel turns 1 to 6 images into walkable 3D and bullet-time video | 1 to 6 input images |
| Muse Voice Transcribe · Meta AI | Voice-agent builders | Meta Muse Voice Transcribe: streaming ASR with diarization and endpointing in one model | 3.1% streaming WER (claimed) |
| GWM Worlds 2 · Runway | Audio & speech teams | Runway GWM Worlds 2: real-time 720p, 24 fps world model with 48 kHz audio | 24 fps @ 720p real-time generation |
Qwen3.8-Max-0902: 2.4T API-only refresh claims #1 on Code Arena
Alibaba refreshed its API-only frontier model with a September 2 snapshot: 2.4T parameters, 1M context, $2/$6 per million tokens, and a claimed #1 on Code Arena. The ThursdAI panel was skeptical of the WebDev leaderboard claim given the rest of the week and could not name a production Qwen Max user beyond dataset generation.
Claude Fable 5.1 & Mythos 5.1: Terminal-Bench 4.0 jumps to 55.8, cache reads 75% cheaper
Anthropic shipped Fable 5.1 and Mythos 5.1 as the same weights with different guardrails. Fable 5.1 scores 55.8 on Terminal-Bench 4.0 (up from 42.0 for Fable 5, versus about 37 for GPT-5.6 Sol) and 52% on the new Terminal-Bench Science, more than double Fable 5. Cache reads drop 75% to $0.25/M while input/output stay at $10/$50 per million tokens. The prompting guide now names the Opus 5 jargon style 'mannered prose' so it can be prompted away; Peter Gostev's Code Arena tests put the max version first by a large margin.
fal H3 Max Turbo: ~97% of H3 Max quality in 1.4 seconds at $0.01 per second
fal's Turbo variant of MiniMax H3 Max keeps about 97% of H3 Max quality, generates a clip in 1.4 seconds, and costs one cent per second at 768p, about 50% cheaper. Alex's live test found it no longer renders copyrighted characters the way H3 Max did.
Gemini 3.8 Flash: third Flash in three weeks, HLE-Verified 54.9, 1M context
Google's third Flash iteration in as many weeks, billed as the reasoning and coding workhorse for Googlers who all build on Antigravity internally. It scores 54.9 on HLE-Verified, keeps a 1M-token context, is about 3x faster, and is priced at $0.75/$3.75 per million tokens until a doubling on January 1, 2027. The Wall Street Journal reported Google scrapped the 3.5 Pro checkpoints because Flash overtook them; Gemini 4 is still in post-training.
Gemini 3.8 Flash Cyber: Fairwind-only cybersecurity variant, CWE-Bench 47.2%
A dedicated cybersecurity variant of Gemini 3.8 Flash, available only to trusted defenders through Google's Fairwind program. The newsletter lists CWE-Bench at 47.2%; no benchmark numbers surfaced during the show, and Alex questioned who will actually get to use it.
Inworld Realtime TTS-2 goes GA: sub-100ms, $25/M characters, #1 on AA Controlled Voice Arena
Inworld's real-time text-to-speech model is generally available with sub-100ms latency at $25 per million characters. It ranks #1 on Artificial Analysis's Controlled Voice Arena and #4 on the Provider arena.
Meta Muse Spark 1.3 ties GPT-5.6 Sol and Grok 4.6 on the AA index, max mode ties Fable 5
Spark 1.3 xhigh scores 61 on the Artificial Analysis Intelligence Index, tying GPT-5.6 Sol and Grok 4.6, and a limited-preview max reasoning mode scores 62, tying Claude Fable 5, the first time Meta has jumped over OpenAI, Microsoft MAI, and Google on that index. Pricing is unchanged at $1.25/$4.25 per million; it runs about 3x faster than Fable and Sol at roughly 4x lower cost than Sol. MRCR long-context jumped from 66% to 98.5% at 1M tokens, and a contributor tier charges $0.10/$0.20 per million if Meta can train on your prompts. Open weights and a mystery 'Watermelon' model are teased as coming soon.
Meta Muse Voice Transcribe: streaming ASR with diarization and endpointing in one model
Meta's streaming speech-to-text model handles transcription, speaker diarization, and endpointing in a single model, claims 3.1% streaming WER, and is API only. It costs about 18 cents per hour, supports a custom dictionary (it transcribes 'ThursdAI' correctly), and powers the live diarized transcript on thursdai.news/live, where Alex says it beats Descript on names and terms.
Microsoft MAI-Transcribe-2: 2.0% WER, 400x real time, $1.67 per 1,000 minutes
Microsoft AI's second transcription model ranks #2 on the Artificial Analysis WER leaderboard at 2.0%, runs about 400x real time, and costs $1.67 per 1,000 minutes, less than half the price of its peers. It ships real-time ASR with diarization the same week as Meta's Muse Voice Transcribe.
OpenAI launches GPT-6 Astra, calls it its most intelligent and aligned model
OpenAI released GPT-6 Astra on September 3, 2026, live during the ThursdAI stream, with president Greg Brockman telling reporters 'welcome to the AGI era.' OpenAI reports 99.9 on ARC-AGI-3, 97.6% on FrontierMath Tier 4, 72.6% on OSWorld 2.0, 92.7 on ScreenSpot Pro and 64.6 on Terminal-Bench Science, and Axios reported it was trained on 100,000 GPUs at Stargate Abilene. API pricing is $10/$50 per million tokens (same as Fable 5.1, 2.5x Sol's promotional price) with a Fast mode at 2x the price, available in the OpenAI API and Amazon Bedrock, rolling out from a limited set of organizations to Plus, Pro, Business and Enterprise. Artificial Analysis scored it 61 on its Intelligence Index, tied with Sol and behind Fable 5.1, but about 70% more token efficient than Sol.
Runway GWM Worlds 2: real-time 720p, 24 fps world model with 48 kHz audio
Runway's second generative world model runs in real time at 720p and 24 fps with open-ended sessions, 48 kHz audio, and generated speech, a big jump over the low-fidelity audio in most world models. LDJ broke it live on the show two days after Runway's previous world model. Research preview.
Runway Solaris: an Interface World Model that generates clickable UIs frame by frame
Solaris generates interfaces as video, so every element in a scene is clickable and draggable: click a lamp and the room lights up, drag shoes onto a person and he wears them. In Runway's own study the generated behavior was preferred 61 to 24 over Opus 5 (71% preferred it over coded UI). Research preview, no public access yet.
Tencent Hy4 preview: 770B/49B Apache 2.0 MoE, Sherry quant shrinks 1.5 TB to 214 GB
Tencent open-sourced a 770B-parameter MoE with 49B active, 1M context, and an Apache 2.0 license. Its Sherry quantization takes the weights from 1.5 TB to 214 GB at 2.38 bits per weight. Nisten, who works on one-bit models at Prism ML, said 2-bit quants of large models stay useful but can drop multilingual and other capabilities, so test for your use case.
World Labs Atlas: omnimodel turns 1 to 6 images into walkable 3D and bullet-time video
Atlas is pretrained from scratch to natively operate in text, images, video, and 3D as an autoregressive diffusion transformer. From 1 to 6 images it generates up to a minute of 1440p camera-controlled video and a 3D reconstruction with no Gaussian splats, and it can reframe real video from new angles, producing bullet-time from three ordinary tripods. LDJ noted quality scales with more camera angles. Partner access only for now.
Kimi K3 (2.8T) runs on CoreWeave Dedicated Inference on GB300 NVL72
CoreWeave added Moonshot's 2.8T-parameter Kimi K3 to Dedicated Inference, running on GB300 NVL72 systems. Alex says it 'purrs like a kitten' and a token promotion is planned on the CoreWeave and W&B accounts.
fal.live: viewer-steered infinite video stream built in a weekend after Twitch and Kick bans
After fal's infinite Rick and Morty 'interdimensional cable' stream was banned from Twitch and then Kick for copyright, the team built its own streaming site over a weekend with a model tuned for continuous generation. Viewers vote on where the anime and other channels go next. Alex credits it with inspiring the thursdai.news/live build.
Muse Code leaves beta with $5 / $20 / $50 plans and a TypeScript SDK preview
Meta's coding agent is out of beta with plans at $5, $20, and $50 per month, a TypeScript SDK preview, and a contributor tier at $0.10/$0.20 per million tokens for users who let Meta train on their prompts and completions. Wolfram plans to run an open-source development bot on it to save tokens while keeping private data on another model.
Anthropic ships background computer use in Claude, Claude Code, and Cowork
Claude can now drive the computer in the background from the Claude app, Claude Code, and Claude Cowork. Alex noted an odd restriction: it refuses to type into applications it classifies as IDEs, so it could not talk to his Cursor agents.
Codex gets GPT-6 Astra: notes across context windows and async questions
With GPT-6 Astra, Codex can keep notes across context windows and search earlier messages and tool output instead of relying on compaction, and it can ask the user a question without stopping work that does not depend on the answer. The feature ships experimentally behind a config.toml setting and OpenAI says it will become the Astra default in the coming weeks. A new Codex harness using Astra completed Mind2Web tasks 1.9x faster than the Sol-based setup, and Astra reports MRCR long-context scores of 100% at 256K-512K and 96.3% at 512K-1M.
Abliteration AI hosts a refusal-free GLM-5.3 at $5/M with only CSAM and self-harm blocked
Abliteration AI released abliterated-model-large-v2, Z.ai's GLM-5.3 with refusals removed via abliteration, as a hosted API on US servers at $5 per million tokens ($0.50 cached). Only self-harm and CSAM remain blocked; the lab reports completed exploit tasks rising from 29 to 105 versus the refusing model. Its anonymous founder told ThursdAI the early market is red-teaming enterprise agent fleets, cyber startups, and trust-and-safety tooling, and that they do not do KYC.
OpenClaw 2.0: native computer use via Cua Driver, cloud fleets, 16,977 PRs
The open-source agent framework's 2.0 release adds native computer use through Cua's driver, cloud fleets, a gateway, and platform support spanning Mac, Linux, Windows, iPhone, iPad, Android, Wear OS, and Docker, after 16,977 pull requests. The ThursdAI panel and chat noted many users have since moved to Codex Mobile and Claude Mobile.
GPT-6 Astra system card: zero honeypot attacks, but reasoning is harder to monitor
OpenAI's system card for GPT-6 Astra reports 2% on its internal computer-use safety benchmark (down from 22% for GPT-5.6 Sol), zero attempts to attack outside the sandbox on Exploit Gym Honeypot, the metric created after the July 2026 swarm incident where Sol attacked roughly half the time, and FrontierCyber results with real zero-days that earned a critical cyber designation. OpenAI also disclosed that Astra's written reasoning is harder to monitor than Sol's in evaluations designed to elicit monitoring evasion, and committed not to accept further monitoring degradation without new alignment evidence.
Don't wait a month — get every week's releases recapped in your inbox each Thursday.
