LLM

13 articles

Kimi K3.1 Pro benchmark wins and losses against Claude Opus 5.5, API pricing for K3.1 Pro and Highspeed, and the open-weight release status

Kimi K3.1 Pro: A Near-Frontier Model You Can Download, Almost

Kimi K3.1 Pro for production AI teams: Artificial Analysis Intelligence Index rank and cost per task, benchmarks against Claude Opus 5.5, $2/$10 pricing, API changes and an open-weight readiness plan.

LLM Production AI

25 Oct, 20267 min read

Gemini 4 Argon benchmark wins and losses, introductory and standard pricing, and the staged access rollout

Gemini 4 Argon: Get Ready Before You Can Call It

Gemini 4 Argon for production AI teams: benchmarks against GPT-6 Astra and Opus 5.5, pricing that doubles after launch, 1M output tokens and a readiness plan.

LLM Production AI

04 Oct, 20265 min read

September 2026 in AI for production teams: frontier prices falling, managed agent runtimes, agent security incidents and new AI laws

AI in September 2026: Cheaper Models, Managed Agents and New Rules

The AI news from September 2026 that matters to teams shipping AI: Claude Opus 5.5, GPT-6 Sol and Luna pricing, OpenAI's Agents API, Plugin4Shell, California's new AI laws and a migration checklist.

LLM Production AI

26 Sep, 20267 min read

State and typed questions go into Jev, typed answers with confidence come out, and application code decides whether to act, confirm or escalate

TypeSafe Jev: A Model That Makes Decisions Instead of Writing Text

TypeSafe's Jev explained for production AI teams: typed Choice, Score and Noul answers, confidence gating, $0.042 per million input tokens, and the caveats in its benchmarks.

LLM Production AI

19 Sep, 20265 min read

Release gate for adopting GPT-6 Astra: candidate, evaluate, shadow traffic, then route only where it wins

GPT-6 Astra in Production: What Changes for AI Engineering Teams

How to evaluate, route and govern OpenAI's GPT-6 Astra in production: pricing, refusal behaviour, reduced reasoning visibility, agent risk and a migration plan.

LLM Production AI

06 Sep, 20263 min read

Router sending requests to Flash-Lite, Gemini 3.8 Flash, or a frontier model by measured task quality

Gemini 3.7 and 3.8 Flash: When the Cheap Tier Gets Good at Coding

Gemini 3.7 and 3.8 Flash for production AI teams: benchmarks, pricing that doubles after December 2026, deprecated sampling parameters and tiered routing.

LLM Cost Optimization

06 Sep, 20264 min read

RAG pipeline from sources through permissioned indexing, retrieval and reranking to a grounded answer

How to Make a RAG System Reliable in Production

A CTO guide to reliable production RAG systems: source quality, retrieval evaluation, access control, freshness, fallbacks, and monitoring.

Infrastructure LLM

18 Aug, 20262 min read

Evaluation pipeline from dataset to quality gates, splitting into release and hold decisions

How to Build an LLM Evaluation Framework for Production

How to build an LLM evaluation framework: golden datasets, offline and online evaluation, LLM-as-judge, release gates and production feedback loops.

Evaluation LLM

11 Aug, 20266 min read

Quality-aware router directing requests to cache, small model, or frontier model to reduce spend

How to Reduce LLM Inference Costs Without Losing Quality

A playbook for reducing LLM inference costs: per-workflow cost profiling, quality-aware routing, semantic caching, prompt compression and infrastructure.

Cost Optimization LLM

29 Mar, 20265 min read

Frozen 4-bit base model with trainable LoRA adapters and a single 80 GB GPU memory budget

Fine-Tuning Llama 3 with QLoRA: A Practical Guide

A practical guide to fine-tuning Llama 3 with QLoRA, covering data preparation, training trade-offs, and production deployment considerations.

LLM Research

20 Mar, 20262 min read

Versioned prompt templates moving through a test suite and canary rollout into production

Prompt Engineering at Scale: Moving Beyond Hacks

Move beyond ad hoc prompts with versioning, testing, observability, and systems that make prompt engineering reliable at scale.

LLM Engineering

10 Mar, 20262 min read

Comparison of Pinecone, Weaviate, Qdrant and pgvector across operations, scale, filtering and cost

Vector Databases Explained: Choosing the Right One for Your AI Application

Compare vector database options for production AI applications, including performance, scalability, cost, and operational trade-offs.

Infrastructure LLM

05 Mar, 20262 min read

Decision split between retrieval for missing knowledge and fine-tuning for wrong behaviour

RAG vs Fine-Tuning: How to Choose the Right Approach for Your LLM

Understand when to use retrieval-augmented generation or fine-tuning, with practical trade-offs for production LLM applications.

LLM Engineering

20 Jan, 20262 min read