AI developer tools · What shipped
For a senior engineer who already reads HN. Real changes in AI developer tools today: releases with version numbers, papers with benchmarks, repos that crossed a threshold worth knowing. Skip hype threads, pre-announcement leaks, and recycled summaries. Always link primary sources.
← See today's latestMeta Muse Code now live, DeepSeek V4 reasoning suite expands
Meta Muse Code (beta)
Meta's terminal agent is shipping today on macOS and Linux.
Muse Code runs Muse Spark 1.2, a model purpose-built inside its own harness so agent and runtime co-optimize as one unit [Quelle: Meta]. The agent carries persistent async workers, replay-safe crash recovery, and bundled skills like /plan for approval gates and /grill for stress testing. Live demos showed iterative kernel writing and profiling across 1,000+ tool calls. Pricing at $0.10/$0.20 per million tokens on contributor tier undercuts mid-tier rivals by roughly 12x.
The catch: performance degrades outside Meta's environment—a coupling tax other vendors must now price.
DeepSeek V4 reasoning models
DeepSeek shipped four model variants targeting agentic workflows this week.
V4-Pro is the 1.6-trillion-parameter flagship with 49B active tokens, designed for advanced reasoning and autonomous agents [Quelle: ZenMux]. V4-Flash offers efficiency with lower latency, 1M context windows, and open-sourced weights. V3.2 introduces large-scale agent training across 1,800+ environments and integrated thinking modes for tool interaction. All models support structured tool calling, accessible via OpenAI and Anthropic interfaces with weights on Hugging Face.
Benchmark parity with o1 reasoning at a fraction of the compute footprint reshapes the frontier cost curve.
Meet Muse Spark 1.2 and Muse Code, the first coding agent from Meta8 hours ago ... Install Muse Code with one command and authenticate in your browser at dev.meta.ai. Available now in beta. These cookbooks are written to help developers get ...developer.meta.com
DeepSeek | ZenMux AI Model Routing49 minutes ago ... 2 is a reasoning-first large language model released by DeepSeek as the official successor to V3. ... The model improves tool use, code generation, and reasoning ...zenmux.ai