Benchmark: Clef 27B Matches Opus 5.5 at a Fraction of the Cost in Gateway Decision Tasks
A benchmark of 22,000 decisions across 3 decision models and 8 LLMs on human-labeled gateway tasks shows Clef 27B reaching 83.2% accuracy at $0.13 per 1K items…
Global AI Digest
-
-
Octop Open-Sourced: A Self-Hosted Multi-User AI Workspace with Per-Account Isolation
Octop, a self-hosted multi-user AI workspace, has been open sourced. Each account gets its own isolated environment with swappable models, storage, memory, and…
-
Codex Auto-review Now Free for All ChatGPT-Signed-In Users
OpenAI Codex introduces Auto-review as a free feature for all users signed in with ChatGPT, letting a second agent monitor the primary agent's actions to block…
-
How LiteLLM Lens Processes Thousands of Agent Traces Beyond Context Window Limits
A behind-the-scenes look at how LiteLLM Lens analyzes thousands of agent traces, some larger than a model's context window, to identify repeated failure pattern…
-
LiteLLM Lens Adds Datasets Feature for Building Agent Evaluation Loops
LiteLLM Lens introduces a datasets feature, enabling a workflow where failing traces are collected into datasets and used to run evaluations on agents.
-
GLM-5.3 Now Available on Amazon Bedrock
Z.ai announces that its GLM-5.3 model is now available on Amazon Bedrock, enabling enterprise users to leverage its coding and agentic capabilities.
-
Cline Introduces Pareto 26.10 Preview Model With 56x Cost Advantage
Cline announces Pareto 26.10 Preview, a routing model that picks the best response from frontier and open models while preserving prompt cache, delivering the s…
-
Reflection AI Introduces Beam: 501B-Parameter Agentic Open Model Coming to Ollama
Reflection AI announced Beam, a 501B total parameter / 23B active agentic open model trained from scratch, with full weights releasing this month; Ollama confir…
-
Command Code Launches First Open Models Family, Explains 'Agr' Name
Ahmad Awais announces Command Code's first open models family and explains the Urdu origin of the name 'Agr', meaning 'if', reflecting the model's probabilistic…
-
Command Code Launches Agr Open Decision Model for Tool Calls and Routing
Command Code launched Agr, an open decision model family optimized for coding agent tool calls and routing decisions, scoring 58.15 on Decision Index 0.2.1 with…
