LG opens K-EXAONE 2.5 as a free API, rivaling DeepSeek V4.1 Flash

Free keys live now for dev and research. Plus Microsoft's APM packages agent setups, and Cognition's SWE-2 Flash runs free in Devin through Nov 30.

Nowline NOV 9 5:00 AM banner

Top AI stories from the last hour

Top AI stories from the last hour

Copy markdown

  • A free API that holds its own vs DeepSeek V4.1 Flash

    LG AI Research opened K-EXAONE 2.5 as a free API at Nvidia AI Day Seoul — grab a key on the Exaone site and build today, no cost for dev or research. Its 24-benchmark suite is up 70% over August and sits level with DeepSeek V4.1 Flash (Automation Bench 74.6 vs 74.9, ClawEval 77.3 vs 75.8, SWE-Bench Pro 59.2 vs 58.4); it's API-only for now.

  • A package manager for your agent setup

    Microsoft's open-source APM treats skills, prompts, hooks and MCP servers like npm dependencies: one apm.yml plus a lockfile with content hashes, compiled into the native files of nine harnesses including Claude Code, Copilot, Cursor, Codex and Gemini. 'apm audit' and 'apm-policy.yml' add CI drift checks and source allow-lists, so a team pins and shares one agent config instead of copying dotfiles.

  • Cognition's fast coder, free in Devin until Nov 30

    SWE-2 Flash runs about 3x faster at one-fifth the price of SWE-2, and it's free in Devin Desktop and the CLI through November 30 for Pro, Max and Teams users. It's a cheaper default for high-volume agent loops where full SWE-2 was overkill.