Cloudflare Clef - Cloudflare's Open-Weight Decision Models for Agent Routing

Microsoft's Microsoft-Decision-1 is a specialized decision-scoring model for engineering teams building routing, classification, and verification logic into their AI systems. Announced on October 9, 2026, it returns calibrated scores for a fixed set of choices rather than generating free-form text, which makes it predictable to drop into production pipelines. It targets teams that already run m...

Saluki 27B is a 2-bit compressed build by Underdog of Alibaba's Qwen3.8-27B that gives local agents a tool-calling model in a 7.89 GB file, beating the full 54 GB original at function calling. Released on October 7, 2026 by Conway Research's Underdog project, an a16z-backed on-device assistant effort, it is licensed under Apache 2.0 and runs on stock llama.cpp — making it a practical agent brai...

Liquid AI Open d1 is a family of open-weight decision models for developers and edge engineers who need fast, structured answers on device instead of token-by-token generation. Released on October 7, 2026, the d1 models do not produce text at all; they score a fixed set of answer options in a single forward pass, which makes them a fit for real-time routing, moderation, and agent control loops....

Claude Haiku 5.5 is Anthropic's smallest and fastest model in the Claude 5.5 family, built for teams that need high-volume, cheapest, low-latency inference. Released on October 7, 2026, it targets the repetitive but important jobs: document summarization, classification, routing, and the helper sub-agents that feed larger models such as Sonnet 5.5 and Opus 5.5. Core Features Prompt-length tiere...

Aleph Alpha Kolibri is a sovereign open-weight MoE model from Aleph Alpha, a German company that released it in Germany on October 3, 2026 under the Apache 2.0 license. It is built for public-sector and regulated-industry teams that must run capable language models on their own infrastructure under EU data rules. Core Features Mixture-of-experts architecture with 78.1B total parameters but only...

Mistral Large 4 (ML4, codename leChonk) is Mistral AI's native multimodal frontier model for enterprises that need a Europe-hosted, high-capability model for reasoning, coding, and document-heavy knowledge work. It entered public preview on Mistral Studio on October 6, 2026. Core Features 1 trillion total parameters with 49 billion active per token, trained from scratch on 3,800 Nvidia Grace Bl...

Grok 4.7 is xAI's frontier model launched on September 21, 2026 for coding, agentic tasks, and knowledge work, giving developers a 500K-token context window plus image input and configurable reasoning. It is the new flagship behind Grok Bot and is served through the xAI API, Cursor, Grok Build, and AWS Bedrock. Core Features 500K context window with text and image inputs and text-only output, a...

Cognition SWE-2 is a coding model for engineering teams that want frontier-level software-engineering performance without paying frontier token prices. Released on September 10, 2026 inside Devin Desktop and CLI, SWE-2 is post-trained from Moonshot AI's Kimi K3 and routed through the Devin product rather than sold as a standalone API. Core Features Reinforcement-learning post-training from Kimi...

Google Gemma 4 26B A4B is an open-weight Mixture-of-Experts language model from Google's Gemma family, built for developers who want near-frontier quality at a price they can run locally. The 26B is total parameters on disk; A4B is the roughly 3.88 billion parameters activated per token, so it behaves like a 4B dense model at inference. Community MLX work pushed straight-in-memory inference abo...

DeepSeek V4.1 Flash is DeepSeek's lightweight multimodal flagship of the new Causal Encoder-Decoder family, built for developers and businesses that need frontier reasoning at low latency and long context. Released through the DeepSeek API and Alibaba Cloud ModelStudio on September 13, 2026, it ships native multimodal vision so it can read charts, documents, and screenshots without a separate m...