Click any tag below to further narrow down your results
Links
Google's new Decisions API lets developers get probability scores for closed-set choices like routing, policy checks, and grading, instead of generating text token by token. It's built on the Gemini Decision 1 model and aimed at fast, cheap control-plane decisions.
- Gemini Decision 1 scores candidate options in a single forward pass, returning probabilities and confidence scores at 150-250 ms end-to-end latency.
- Pricing is $0.10 per 1M input tokens with zero output tokens billed, and Google says it's nearly 5x cheaper than Gemini 3.5 Flash-Lite with the Interactions API on the shared benchmark suite.
- Each request can carry one multimodal input (text, images, audio, video, PDFs up to 256K tokens) and up to 1,000 questions, with yes/no, multiple-choice, and rubric-score types that return confidence values so ambiguous cases can go to a fallback or human.
Drex Max Finance is a 27B multimodal model from Nace.AI, tuned for banking, accounting, payments, insurance, and compliance work. It takes text, JSON, or document images like invoices and statements, then returns calibrated probabilities for each answer option in one forward pass, with no text generation.
- It outputs a calibrated probability for every option on yes/no, multiple choice, or ordinal questions, in a single forward pass rather than generating text.
- It accepts documents as input (invoices, receipts, statements) alongside structured data and plain text.
- It supports contexts up to 262K tokens.
Thinking Machines has released Inkling 2 Small, a 594B-parameter mixture-of-experts model with 17B active parameters that accepts image input. It's available now through their Tinker platform, with full weights promised under Apache 2.0 after external safety testing.
- The model scores 31 on the Artificial Analysis Intelligence Index, up from the company's previous generation, with gains credited to more efficient pre-training and a large unified RL run.
- Coding results are mixed rather than dominant: it places second on several benchmarks like FrontierCode v2.0 (25.4%) and DeepSWE (72.1%), behind DeepSeek-V4.1-Flash on most rows.
- On a self-run autoresearch test, the model autonomously fine-tuned Qwen3.5-4B over 4.5 hours for $426 in Tinker spend and more than doubled its forecasting score.
This arXiv paper argues that models trained on different data types can converge on a shared representation, and that it can be proven under specific conditions, not just hoped for as scale grows. Its central finding is that the "good views" used to train joint-embedding predictive architectures, not model size alone, are what drive that alignment.
- The authors prove that JEPAs trained separately on different modalities learn the same representation up to a rotation, provided their positive pairs are two views of the same scene.
- Next-token prediction is framed as a JEPA with "bad views" (parts of one document), which they argue explains why language models align with vision only through scale.
- A 109M-parameter text JEPA aligns with DINOv2 and CLIP image encoders better than every language model tested up to 72B parameters, and needs 42 times fewer samples than next-token prediction to reach the same alignment.
This is the sign-in page for Liquid AI's chat playground, where you can talk to the company's own family of models. The lineup spans text, vision, and voice, with the pitch being strong performance at much lower compute cost than typical frontier models.
- The lineup includes LFM2.5-24B-A2B and LFM3-1.7B for text, LFM3-VL-2B for vision, and LFM3 Omni 2B and LFM3 Audio 2B for voice.
- There is also LFM2.5-1.2B-JP, a text model aimed at Japanese.
- The page offers no benchmarks or pricing, so the efficiency claim is unverified from this text alone.
Liquid AI has released LFM3, a family of roughly 2B-parameter open-weight models covering text, vision, audio, omni, and decision tasks on one shared backbone. The pitch is strong capability per parameter that runs fast on phones, laptops, and single GPUs, and can be fine-tuned locally.
- The LFM3-1.7B text model has a 256K-token context window and leads several agentic and tool-use benchmarks, including BFCL and ToolSandbox, while beating Qwen3.5-9B on WebArena-Verified and AutomationBench.
- On-device, it decodes 1.8 to 2.8 times faster than the next-best model tested and needs only 1.6 GB of KV cache for a full 128K context, versus 4.3 GB for Qwen3.5-4B.
- The architecture mixes gated short convolutions, sliding-window attention, and global attention, and the team reports about 25% faster inference than LFM2 at 8K context.
DigUp is a free, open-source desktop app that lets you search files on your computer by describing what you're looking for in plain language—finding photos, PDFs, videos, and audio by meaning rather than filename. It uses Google DeepMind's EmbeddingGemma 2 model to understand text, images, audio, and video across 100+ languages, and everything runs locally on your machine with no cloud uploads or tracking.
- Searches across multiple file types (images, PDFs, documents, audio, video) in a single query, matching both semantic meaning and exact strings like invoice numbers or error codes
- The 740M-parameter EmbeddingGemma 2 model runs entirely on your computer—all indexing stays local, with only one-time model download and optional daily update checks
- Includes code search that can find functions by what they do ("retry a failed call with backoff") or by name, opening results directly in your editor at the correct line
- Initial indexing takes minutes to hours depending on file volume, then new files become searchable within seconds
TypeSafe released Jev 1.14.0 with a split approach: stable improvements for production use and experimental "danger zone" features (image support, bounding box type, longer context) for developers willing to accept less reliability. The company is also sharing internal intelligence metrics while acknowledging their limitations.
- Image support is now composable across Jev's JSON interface (state, instructions, criteria), though multimodal performance is weaker than text-only and requires careful prompt decomposition.
- New "box" type returns bounding boxes without type errors, representing the first Jev type designed beyond binary decisions and signaling exploration into broader System One Model capabilities.
- Jev supports longer context lengths than previously enabled in production; developers can opt in via a danger zone flag, but TypeSafe intentionally limits context by default because reliability degrades at scale.
- Internal benchmarking data is being shared for the first time, limited to open-weight models only, with explicit caveats that the methodology will evolve and developers should run their own evals.
Chandra OCR 2 is Datalab's optical character recognition model that converts documents to markdown, HTML, or JSON while preserving layout. It scores 85.8% on the olmOCR benchmark and supports 90+ languages with strong performance on tables, math, handwriting, and complex layouts.
- Chandra 2 achieves 85.8% on olmOCR benchmark (state-of-art), a 12% improvement over Chandra 1's 77.8% multilingual score, with particular strength in math, tables, and complex layouts
- Available through multiple deployment methods: HuggingFace Transformers, vLLM, SGLang, and Docker, with quickstart CLI tools for local use
- Handles 90+ languages, extracts images with captions, reconstructs forms accurately including checkboxes, and supports handwriting recognition across diverse document types
The article introduces Inkling-Small, a Mixture-of-Experts transformer with 276 B total and 12 B active parameters that matches full Inkling performance at one-quarter the size. It handles text, images, and audio with a 1 M-token context window and lets users trade compute for thinking effort. The open-weight model outperforms peers on tool use, coding, reasoning, and multimodal benchmarks and is available for fine-tuning and chat.
- Inkling-Small matches or beats full Inkling performance (e.g., 31.6% vs 29.7% on Humanity's Last Exam) while using only 12B active params out of 276B total—roughly a quarter the compute cost.
- It's natively multimodal (text, images via 40×40 patches, audio via dMel spectrograms) with a 1M-token context window and adjustable "thinking effort" for cost/performance tradeoffs.
- Safety numbers slightly lag full Inkling (71.6% vs 78.0% harmful-query refusal on FORTRESS) despite reusing the same safety recipe.
- Open-weight and available now for fine-tuning on Tinker and chat testing in the Tinker Playground.
ChatGPT Images 2.0 uses a new model that overcomes diffusion-driven spelling errors and renders small text, icons, and UI elements cleanly at up to 2K resolution. It adds web search, multi-image prompts, and support for non-Latin scripts, with knowledge capped at December 2025. All ChatGPT and Codex users get access Tuesday, with paid tiers unlocking higher-quality outputs and an API.
- ChatGPT's new Images 2.0 model renders text, icons, and UI elements accurately at up to 2K resolution, fixing the garbled-text problem that plagued earlier diffusion-based image generators
- It adds "thinking" features like web search, multi-image generation per prompt, and self-verification of outputs, plus cleaner rendering of non-Latin scripts (Japanese, Korean, Hindi, Bengali)
- Training data is capped at December 2025, so it may fumble very recent events
- Rolling out Tuesday to all ChatGPT and Codex users, with paid tiers getting higher-quality output and a new gpt-image-2 API priced by resolution and fidelity