SShortSingh.
Back to feed

Mistral Small 3.2 (24B) Released with Improved Function Calling and 128K Context

0
·1 views

Mistral has released Mistral Small 3.2, a 24-billion-parameter open-weight model now available on Hugging Face under an Apache 2.0 license. The update significantly improves function calling reliability by reworking tool schema handling, fixing issues with optional parameter parsing that caused hallucinations in version 3.1. The model supports a 128K context window with sliding attention enabled by default in vLLM 0.6.3 and above, improving throughput for long-context workloads. Benchmark testing places it within 3–5% of GPT-4o mini on most reasoning tasks, with multi-turn tool-use accuracy rising from 71% in Small 3.1 to 78% in the new version. Quantized builds, including 4-bit GGUF variants, are already available, making the model accessible for local development on hardware such as Apple's M2 Ultra Mac Studio.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Log in to join the discussion and vote.

Log in

Related stories

0
ProgrammingHacker News ·

German startup Isar Aerospace becomes first to reach orbit from European soil

German private aerospace company Isar Aerospace has made history by successfully launching its Spectrum rocket into orbit from European territory. The launch took place at Andøya Spaceport in Norway, marking a significant milestone for the European commercial space industry. This achievement represents the first time a privately developed European rocket has reached orbit from a launch site on European soil. The feat signals a growing capability for Europe to compete in the commercial launch market independently of non-European providers.

0
ProgrammingDEV Community ·

Why AI Confidence Scores Can Be Dangerously Misleading in High-Stakes Systems

A technical analysis published on DEV Community argues that confidence signals in large language models (LLMs) are structurally disconnected from real-world reliability, posing serious risks in high-stakes fields like healthcare, finance, and software engineering. LLMs generate text based on token probability distributions and do not natively verify the truth of their outputs, meaning they can hallucinate with the same authoritative tone used for accurate information. The article distinguishes between token probability, claim reliability, decision confidence, and action safety, warning engineers against treating model outputs as direct proxies for correctness. Beyond model-level issues, production systems face additional failure points such as stale data indexes, API errors, and ambiguous user inputs that compound uncertainty. The author concludes that addressing AI hallucination requires system-level uncertainty controls, not just model fine-tuning, and that confidence must actively govern what an AI system is permitted to do.

0
ProgrammingHacker News ·

Terence Tao Demonstrates Finite-Time Blowup in Averaged Navier-Stokes Model

Renowned mathematician Terence Tao published a research post in February 2014 exploring finite-time blowup behavior in a modified version of the three-dimensional Navier-Stokes equations. The work examines an averaged variant of the equations rather than the original formulation, which remains one of mathematics' unsolved Millennium Prize Problems. Tao's findings show that solutions to this averaged model can break down in finite time, suggesting potential pathways toward understanding instability in fluid dynamics. While the result does not directly solve the full Navier-Stokes regularity problem, it provides theoretical insight into how blowup might occur. The post appeared on Tao's widely followed mathematics blog and drew attention from the broader mathematical and computer science community.

0
ProgrammingHacker News ·

Isar Aerospace Attempts Orbital Launch

Isar Aerospace, a European private rocket company, conducted a launch attempt targeting orbital flight. The mission was documented and shared via video on YouTube. Isar Aerospace is among a new wave of European commercial launch startups aiming to provide independent access to space. The launch represents a significant milestone for the company and for Europe's growing private space sector.

Mistral Small 3.2 (24B) Released with Improved Function Calling and 128K Context · ShortSingh