ModelsRadar
Thinking Machines Lab launches Inkling 2 Small on Tinker with closed weights
The model is available for inference and fine-tuning; the planned Apache 2.0 weight release depends on external safety testing.
Radar
Short notes on AI releases and changes, linked to original sources.
43 articles · Page 2 / 3
Codex & codeRadar
Mattt launches AnyDecisionModel to return typed decisions in Swift
The application programming interface returns choices and probabilities without generated prose; session answers do not become context for later questions.
ModelsRadar
Liquid AI launches six LFM3 models with different input and output modalities
Text, vision, audio, Omni, and decision variants share a backbone; image, video, and speech support depends on the selected model.
ToolsRadar
David Corvoysier publishes a model support lookup for AWS Trainium
The catalog supplies reproduction scripts, software versions, peak device memory and total run time including compilation; passing tests generate just 32 tokens.
AgentsRadar
ThinkingBox uses database-state rewards to train Qwen3.8-27B for 50 steps
After 50 steps of Group Relative Policy Optimization, Qwen3.8-27B passed all 20 attempts on 107 tasks, up from 38; at-least-once coverage fell from 453 to 446.
ModelsRadar
Mistral Large 4 adds text-and-image inference with three reasoning levels
Transformers supports low, medium, and high reasoning effort; none still selects high. The weight license restricts model-as-a-service offerings.
ModelsRadar
Project ProtoStar launches open challenges to test scientific data reuse
The initiative seeks to make scientific datasets, models, and code reusable, starting with urban climate forecasts compared against historical averages and linear trends.
ModelsRadar
Ai2 introduces ToyBlocks to study how robot skills transfer
A simulated arm and tabletop blocks let researchers vary training demonstrations and evaluate performance on tasks held out of training.
ModelsRadar
Open ASR Leaderboard adds Egyptian and Nigerian English evaluations
The automatic speech recognition benchmark adds 1.34 hours of Egyptian-accented English and 1.55 hours of Nigerian-accented English, both conversational and spontaneous.
ModelsRadar
NVIDIA adds four Cosmos checkpoints for controlled simulation video
The Cosmos 3 variants generate driving observations or action-conditioned video, including synchronized camera observations and LiDAR outputs for 3D sensing.
ToolsRadar
Phonon generates speech offline under model-specific licenses
Phonon converts text into speech on-device; its English-only 31m model uses Apache 2.0, while the multilingual 40m model has noncommercial terms.
Codex & codeRadar
Dharma Field Extractor converts document images into typed fields
The 2-billion-parameter model returns structured JavaScript Object Notation (JSON), using a supplied schema or extracting available fields.
ToolsRadar
Palette separates slide design planning from code generation
IBM’s slide-building approach trains a model on a visual design language, then checks rendered slides and repairs layout errors.
Codex & codeRadar
CWE-bench v2 adds 100 repository security repair tasks
Agents must fix vulnerabilities without breaking functionality; deterministic checks score tasks covering 91 Common Weakness Enumeration categories.
Codex & codeRadar
Transformers MIC shrinks models for fast integration tests
Model Integration Contracts pair reduced models with runnable examples and reviewed expected outputs to check integration on standard computer processors.
ModelsRadar
IBM updates Granite Switch speech and decision components
Granite Switch 4.1.1 keeps the base model unchanged and updates speech recognition across 10 languages with punctuation and capitalization.
AgentsRadar
Harness Hub releases adapters trained inside coding-agent harnesses
Each adapter targets a model-and-harness combination; held-out smoke evaluations report fewer tool calls while maintaining or improving accuracy.
ModelsRadar
LeRobot adds dataset streaming and live policy instructions
Version 0.7.0 trains without a full dataset download; questions and autonomous goals require a text-generating policy or an external planner.
AgentsRadar
Hermes Agent adds browser access to its Desktop workspace
Version 0.23.0 gives each session one gateway owner; interfaces attach as clients, and browser access remains disabled by default.
ClaudeRadar
Claude Code runs hooks before and after tool execution
PreToolUse can block a tool call before execution; PostToolUse runs after a successful call and cannot prevent that completed action.