Tag
Kyutai Labs announces that their audio-to-MIDI model MuScriptor now detects tempo, allowing direct drag-and-drop of MIDI into a DAW without manual tempo matching.
Higgsfield announces unlimited Seedance usage for 11 days, with 4 days of Seedance 2.0 in 4K and 7 days of any other Seedance model, ahead of the upcoming Seedance 2.5 release.
DeepSeek V4 Flash High ranks 7th on the overall leaderboard in the frontend code arena with a score of 1586, 3rd among open-source models, a significant improvement over the preview version.
DeepSeek released V4-Flash-0731, an AI model that surpasses Fable-5, Sol, and Kimi-K3 on a chess benchmark.
Nova versão do DS v4 flash 0731 promete grande melhoria por preço baixo, com desempenho superior ao GLM 5.1 em testes caseiros, embora haja desconfiança sobre os benchmarks.
DeepSeek quietly updated its changelog with a V4-Flash upgrade, boosting its Terminal-Bench score to 82.7, a +25.8 leap from the April preview. It is currently API-only, with open weights coming soon.
DeepSeek 宣布更新了 DeepSeek-V4-Flash，并预告 DeepSeek-V4-Pro 的正式发布将很快到来。
Sam Altman announces major price cuts for GPT-5.6 models: an 80% drop for Luna, a 20% drop for Terra, and a new Fast mode for Sol in the API.
Pangram Labs announces Pangram 4, its most powerful AI detector yet, with significantly reduced false positive and false negative rates, robust detection across frontier models, and new image scan features.
Kimi Code releases Kimi K3-256k, a 256k-context version of its flagship K3 coding model, offering reduced quota consumption while maintaining similar performance for most tasks.
Laguna s2.1 launched with impressive benchmarks but suffered from looping and tool usage issues; recent updates may have fixed them, and users are asking if it is now stable.
Muse Spark 1.1 has improved its Text Arena ranking, now close to Fable 5.
xAI announces upcoming releases of Grok 4.6 in two weeks and Grok 4.7 in four weeks.
Google has deprecated temperature, top_p, and top_k parameters starting with Gemini 3.6 Flash and 3.5 Flash-Lite models, as detailed in their developer guide.
Anthropic updated Claude's voice mode to support Opus, Sonnet, and Haiku models, and integrated with apps like Gmail, Slack, Canva, and Notion. Multilingual support in beta.
Kimi-K3 is not yet outperforming Fable, but it is showing significant progress and closing the gap.
Claude Fable 5 will be included in all Max and Team Premium plans starting July 20, with 50% of limits. Pro and Team Standard users get a one-time $100 credit and access via usage credits.
Alibaba's Qwen AI model is being updated, as hinted by a new announcement from the official Qwen account.
This thread presents research on whether new facts can be added to an LLM's weights without breaking the model, and finds that it breaks unexpectedly, making compressed KV caches and in-context learning more promising for continual learning.
After testing Kimi 3, the user found its performance far exceeds the previous generation Kimi 2.7, calling it a 'crushing' victory, and mentioned that a demo is already available on the official website.