Vannarot Roeung
Recent posts by Vannarot Roeung
Qwen-Audio-3.1 Price Cuts: Where the 85% Actually Went
Alibaba sold Qwen-Audio-3.1 with 70 to 95% price cuts. The biggest verifiable cut is on the older 3.0 Realtime Flash, whose audio output fell from $15 to $1.87 per million tokens.
Claude Opus 5.5: Cheaper, But Not a Drop-In Swap
Anthropic cut Opus pricing 20% and took the number one spot on the Artificial Analysis index. The line that will actually change your bill is the effort default, which dropped from high to medium.
MiniMax H3 VAE Is 2x Faster, If You Set the Flag
ComfyUI cut a MiniMax H3 encode and decode round trip from 24.3 to 12.7 seconds. Only one of the three changes is on by default. The other two need a launch flag and a file download.
How to Clean Up Bad Audio With AI in 2026
You can rescue a badly recorded interview in about fifteen minutes using free tools, and the hard part is not which tool you pick. It is the order you run them in. AI speech enhancers are destructive, and stacking two of them is how a usable take becomes a silent waveform.
Step Code's 80.9% Is Three Tasks Over Claude Code
StepFun's Step Code claims 80.9% on Terminal-Bench 2.1. That is 72 of 89 tasks, run once, on a model StepFun has not named, and it is not on the official leaderboard.
Transformers Now Runs GGUF Quants: Benchmarks and Limits
Hugging Face shipped direct GGUF support in Transformers on 22 September 2026. The benchmarks show near-parity with llama.cpp, but a footnote under the chart and four hard constraints decide whether it is usable for you today.
Ming-Image-0.1 Is 6B on Paper, 26B on Disk
Ant Group's inclusionAI released Ming-Image-0.1-Design under MIT: top of the open-weights UI/UX design leaderboard at 1082 Elo, with native RGBA output and a companion model that splits a finished design back into editable layers. The 6B label describes one component of a 52.88 GB package.
MiMo-V2.6-Pro vs Grok 4.7: Same Score, 24x the Bill
Xiaomi open-sourced MiMo-V2.6-Pro under MIT on 21 September and it scored 46 on the Artificial Analysis Intelligence Index. Grok 4.7 launched the same day and scored 46 too. Running that identical evaluation cost $206.66 on one and $4,967.35 on the other.
Snapchat Chat to Song vs Suno: What You Can Keep
Snap added five generative AI features to Lens+ on 21 September 2026, and Chat to Song turns a text message into music. Snap has published no model, no download path and no commercial-use terms for the output.
ComfyUI 0.37 Adds MoGe 3 and Finishes Its 3D Pipeline
ComfyUI shipped v0.37.0 on 21 September with MoGe 3 running as a native node. The node is not the story: it is the last missing piece of a 3D pipeline assembled across three releases since 9 September. Also, the newest version is v0.37.1, and it has no release notes at all.
Hemmingway-1 Is an Email Model, Not a Story Model
Altworld released Hemmingway-1 on 20 September: 27B parameters, Apache-2.0 weights, fine-tuned from Qwen3.8-27B. It tops two of its own benchmarks, places third on the one it does not own, and comes last of seven on both story columns. The numbers were published only inside chart images.
Qwen-Image-2.1 Trades Apache 2.0 for a Research License
Alibaba's Qwen team published Qwen-Image-2.1 on September 20, 2026: 7B parameters, native RGBA output, up to 10 reference images, and day-0 support across Diffusers, ComfyUI and SGLang. It is also the first Qwen-Image release you are not allowed to sell work from.
Qwen3.8-LiveTranslate: Twice the Price, 30% Cheaper
Alibaba released Qwen3.8-LiveTranslate on 19 September 2026, cutting average lag from 2.8 to 2.3 seconds across 60 languages. It charges more than twice Gemini's per-token price for audio and still costs about 30% less per hour. The number that decides your bill is not in the headline.
Six Open Jev Clones, Four Different Jev Scores
TypeSafe opened access to Jev on 15 September. Within 48 hours six open reproductions had shipped, each claiming to match it. Four independent evaluations put Jev's own Banking77 accuracy at 87.0%, 83.2%, 77.8% and 76.3%.
OpenClaw 2026.9.5: Big Release, Dark Stable Channel
OpenClaw shipped v2026.9.5 on 19 September with atomic updates and plugin hot reload. The continuity promise is thinnest in two places: an open hot-reload regression, and an extended-stable channel that published nothing.
Meta Muse Connectors: The Path Meta Doesn't Review
Meta opened the Muse Connector Platform to developers on 18 September 2026, with a three-step review and a Stripe Link payment rail. It governs one of the two ways a service reaches Muse. Custom connectors, which Muse writes itself against any public API, are not reviewed at all.
Claude Code Reads AGENTS.md, With Four Conditions
Claude Code v2.1.277 can read AGENTS.md as a project's instruction file, closing the last conspicuous gap in a standard used by 60,000+ repositories. The documentation tells a more conditional story than the changelog does.
Grok Voice Transcribe 2: 2.3% WER at $0.10 an Hour
SpaceXAI shipped Grok Voice Transcribe 2.0 with two claims stacked on top of each other: twice as accurate as version 1.0, and first among 32 streaming models. Both are defensible. Neither is the reason the release matters.
ZCode Silently Uploads Your Entire .git Folder
A reverse-engineering writeup says ZCode packages your entire working directory, including the .git object store, LFS cache and reflogs, and uploads it to Alibaba Cloud while you are signed in. Here is what was found, and how four coding agents compare on what leaves your machine.
Qwen3.8-Omni-Flash Reads Video at a Fifth of Gemini
Alibaba's new omni model takes text, images, audio and video at $0.15 per million input tokens, exactly a fifth of Gemini 3.8 Flash. The number that decides your bill is not on the pricing page.
Bend 2 Blocks AI Bugs With Proof. Laws Are the Work
Victor Taelin's Bend 2 makes your AI supply a machine-checked proof that your rules still hold. The proof is real. It moves the work into specification, which is the part people are worst at.
How to Publish an AI-Narrated Audiobook in 2026
The two gates that reject AI-narrated audiobooks have nothing to do with how the narration sounds. A seven-step walkthrough from distribution route to a master that meets the published spec, with the FFmpeg commands for every measurement.
Cactus Needle 3: 8 MB Tool Calling, With Conditions
Cactus released Needle 3 on 17 September, an Apache 2.0 automation model in an 8 to 29 MB binary claiming DeepSeek V4 Flash parity. The claim holds on a 200-row benchmark after fine-tuning. The first public outside test scored it 20.4%.
Wispr Canto: 3.4% WER Overall, 21.4% on Short Dictation
Wispr published Canto on 17 September 2026 with a 3.4% word error rate on real dictations, the lowest of six models. Its own charts also show 21.4% on short dictations, where three models tie exactly.