Signing you in...

Please wait while we verify your authentication

Article · Tuesday, September 15, 2026

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.

By Marius BongartsTech66 editions
← See today's latest
Editions
8 / 66
Generated by AI overnight from public sources, refreshed daily.
AI developer tools · What shipped
Tuesday, September 15, 2026
AI developer tools · What shipped

Fable 5.1 cost holds; Mercury 2.5 lands; Visual Studio ships C++ AI

1 min read

Mercury 2.5 diffusion model

Inception Labs shipped Mercury 2.5 with 40% better reasoning at frontier speed.

The diffusion-language model hits 1,107 tokens per second on NVIDIA GPUs while supporting 260K context and pricing at $0.20 per million input tokens ($0.04 at launch) [Quelle: Inception Labs]. Production deployments span search agents, voice apps (170ms to first token), and coding assistants. Mercury Router, now in preview, routes requests across open and closed models for quality-speed-cost optimization.

Early adopters are testing whether diffusion beats transformer latency for real workloads.

Visual Studio C++ + Copilot

Microsoft is weaving AI into C++ workflows at the language-server layer.

Visual Studio 2026 pairs C++-specific code understanding with GitHub Copilot for navigation, modernization, refactoring, and performance tuning [Quelle: Microsoft]. MSVC Build Tools v14.52 (November 2026) adds the /std:c++23 switch and experimental constexpr cmath. Copilot CLI now supports a C++ language-server plugin for semantic intelligence beyond terminal grep.

Systems engineers get richer IDE signals; agents get better compile-and-debug loops.

Fable 5.1 pricing endures

Yesterday's cost advantage holds across all cloud platforms.

Claude Fable 5.1's 75% cache-read cut ($0.25 per million tokens) and 25–45% agentic workload savings remain locked in on Claude API, AWS, Google Cloud, and Azure [Quelle: Anthropic]. Terminal-Bench 4.0 holds steady at 55.8–60.9% across task types; Humanity's Last Exam with tools reaches 65.0%. Mythos 5.1 uses identical internals for vetted teams with 60% fewer false-positive safeguards.

No model has closed the gap on cost-per-reasoning-task since the September 10 drop.

Sources
Introducing Claude Fable 5.1 and Claude Mythos 5.1 - Anthropic
Introducing Claude Fable 5.1 and Claude Mythos 5.1 - Anthropic
7 hours ago ... ... AI experts, testing whether the models could match human specialists' performance. Mythos 5.1's capabilities are greater than those of Mythos 5. However ...
anthropic.com
AI Summary

Anthropic released Claude Fable 5.1 and Claude Mythos 5.1, with Fable 5.1 now generally available. Fable 5.1 costs approximately 25% less than Fable 5 for typical workloads and up to 45% less for highly agentic work, achieved through 75% reduced pricing on cache reads. The model shows performance improvements across multiple benchmarks: Terminal-Bench-Science 0.1 (52.6% vs 24.7%), Terminal-Bench 4.0 (55.8% vs 42.0%), CursorBench 3.2.0 (73.4% vs 70.5%), and Humanity's Last Exam (65.0% with tools vs 63.8% for Fable 5), while maintaining reduced false positives in cybersecurity safeguards (60% fewer than previous versions). Claude Mythos 5.1, restricted to vetted organizations, demonstrated advanced scientific capabilities including designing protein binders with hit rates near 50% (compared to typical 10-15%), creating high-resolution Venus elevation maps, and optimizing GPU kernels for deep learning models with up to 2.5x speedups.

Visit source
Introducing Mercury 2.5 - Inception Labs
Introducing Mercury 2.5 - Inception Labs
3 hours ago ... Since Mercury 2's launch, we've watched Inception advance diffusion-based language models further on NVIDIA AI infrastructure. ... tools, route requests, ...
inceptionlabs.ai
AI Summary

Inception Labs released Mercury 2.5, a diffusion-based language model offering 40% improvement in intelligence over Mercury 2 while maintaining low latency and cost. The model achieves 1,107 tokens per second on NVIDIA GPUs, supports 260K context length, and is priced at $0.20 per million input tokens and $0.75 per million output tokens (80% discount at launch: $0.04/$0.15). Mercury 2.5 is positioned as comparable to cost-optimized frontier models like GPT-4o mini and Claude Haiku, with production deployments across search agents, voice applications (170ms time-to-first-token), and coding assistants. The company also previewed Mercury Voice and Mercury Router, with the latter routing requests across open and closed models for optimal quality-speed-cost tradeoffs. Mercury 2.5 is available through Inception API, Baseten, and OpenRouter with enterprise deployment options.

Visit source
C++ in the Age of AI: Visual Studio at CppCon 2026 - C++ Team Blog
C++ in the Age of AI: Visual Studio at CppCon 2026 - C++ Team Blog
9 hours ago ... MSVC, language conformance, and runtime performance. Copy link. The latest Microsoft C++ (MSVC) Build Tools releases include continued C++23 conformance work ...
devblogs.microsoft.com
AI Summary

Visual Studio 2026 is integrating AI-assisted C++ workflows combining C++-specific code understanding with GitHub Copilot for code navigation, modernization, refactoring, debugging, and performance optimization. MSVC Build Tools v14.52, shipping November 2026, will include the /std:c++23 switch and achieve C++23 completeness, with constexpr cmath implementation available as an experimental opt-in feature. Copilot CLI, built on the modern Copilot SDK shared with Visual Studio and VS Code, now supports a C++ language server plugin for richer code navigation, diagnostics, and semantic intelligence, moving beyond grep-style searches in terminal workflows.

Visit source
Compiled overnight by MorningMail.aiDelivered at 05:10 AM