Tag
A developer demonstrates adding vision capabilities to the GLM language model, showcasing a significant multimodal extension.
Google updated Gemma 4's chat templates with major fixes to tool calling, reduced laziness, enabled Flash Attention 4 on Hopper GPUs, and released an interactive vision guide. The updates are available on Hugging Face.
Google Gemma is rolling out significant improvements to Gemma 4, driven by community feedback and contributions, as detailed in a thread.
The tweet praises SAM3.1 for improved segmentation with faint borders and better tail coverage, leading to more accurate length measurements.
Kimi K3 model is nearing completion, suggesting an upcoming release.
Sam Altman announces that GPT-5.6 sol is half the price and roughly twice as token efficient as fable for many tasks, with plans to deliver at one-quarter of the price.
Devin Fusion now integrates Fable 5, which surprisingly offers lower cost per task than Opus 4.8 due to efficiency improvements in delegation and reasoning.
Sam Altman announces that GPT 5.6 Sol will remain available in ChatGPT subscriptions (Go, Plus, Pro) until a better model is shipped, providing clarity on the model's availability.
OpenAI announces GPT-5.6, a major step forward for health intelligence with stronger performance and 25x lower cost for GPT-5.6 Luna compared to GPT-5.5, making advanced models more accessible globally.
GPT-5.6 achieves a breakthrough by solving a previously unsolved problem, marking a significant advancement in AI capabilities.
MiMo v2.5 is praised for its impressive token generation speed in OpenCode, suggesting it's an underrated model update.
Elon Musk states that Grok is making progress.
OpenAI has made GPT-Realtime-2.1-mini available via its API, providing a cost-efficient option for real-time applications.
Ahmad Osman highlights improvements in Tencent Hy3 over its preview version and compares it to GLM 5.2, which is twice its size, while sharing a prediction about running similar intelligence on an RTX 5090.
Openai is reportedly preparing GPT-5.6 for release next week, with three tiers (Sol, Terra, Luna) and a new reasoning-effort control slider in the Codex app, potentially challenging Anthropic's Fable 5.
Meta is reportedly preparing an update to its flagship AI model Muse Spark.
Third-party tests show that after returning, the Fable 5 model's performance dropped significantly on the BridgeBench benchmark. Debugging score fell from 86.2 to 25.9, Refactoring from 73.6 to 38.4, Hallucination from 75.9 to 61.7. It is speculated that the new safety guardrails caused many tasks to be handed off to Opus 4.8.
Universal-3.5 Pro improves native code switching, diarization, and adds more languages, enhancing speech recognition capabilities.
Anthropic is redeploying Claude Fable 5 globally with updated classifiers to block cybersecurity tasks, though routine coding may trigger fallback to Opus.
Claude Sonnet 5 has appeared in the latest Claude Code CLI build, suggesting a new model release or update from Anthropic.