OPEN WEIGHTS

Kimi Linear explained: Moonshot's KDA hybrid linear attention beat full attention with 75% less KV cache and up to 6x faster decoding at 1M context.

Kimi K3 open weights are on HuggingFace: 2.8T parameters, 1M context, native MXFP4 and a ~1.63TB download. The specs, license terms and HN reaction.

Where to access Kimi K3: Moonshot's API at $3 input and $15 output per million tokens, OpenRouter hosts, inference clouds, and open weights.

Kimi K3 is Moonshot's 2.8T-parameter model with a 1M-token context and Delta Attention. Pricing, HN reactions and where it fits among frontier models.

GLM 5.2 and the AI margin collapse thesis: why open-weights models could squeeze frontier lab margins, where HN pushes back, and what it means for model choice.

Run GLM-5.2 locally with Unsloth dynamic quantization: hardware math for a 256GB Mac or 24GB GPU, quantization tradeoffs and lessons from the HN thread.

GLM-5.2 is an MIT-licensed open-weights coding model. Cheapest routes: OpenRouter hosts from about $0.36 per million input tokens, Z.ai's API, or self-host.

GPT-5.5 hallucinates 86% of the time on AA-Omniscience when it does not know, versus 28% for MIT-licensed GLM-5.2. What the metric measures and why it matters.

Model routers turn having no single winning model into an advantage: how routing works, why open weights and neoclouds make it cheap, and the counter-argument.

Mellum2 developer guide: JetBrains' open-source 12B MoE coding model with 2.5B active parameters. How to run it locally and where it fits in your coding stack.

AI model routing and orchestration decide which model handles which task. How routing differs from orchestration, and practical ways to cut your model spend.

DeepSeek V4 economics for agentic coding: worked cost math, the Flash vs Pro split, and when to route coding work to DeepSeek instead of a frontier model.

GLM-5.2 cost math: when Z.ai's open-weights coding model actually saves money, with per-token pricing, a worked cost-per-task example and a decision guide.

GLM-5.2 vs DeepSeek V4 vs Qwen3 vs Kimi K3: open-weights coding models compared on benchmarks, pricing, context windows, self-host footprint and best picks.

Self-hosting open-weights models like GLM-5.2, DeepSeek V4 or Llama: the break-even math against per-token API prices, with GPU costs and the hidden ops bill.

Fable 5 vs DeepSeek V4 on cost and quality: a 178x output price gap, a real quality difference, and the task split most production teams end up running.

June 10, 2026 AI dev tools roundup - the Fable 5 deadline, PgDog funding, a banking prompt injection, DiffusionGemma speed gains, and what to do this week.

A first-hand visit to DeepSeek HQ reveals something more interesting than benchmark scores: a 300-person company that treats AI as infrastructure, not eschatology - and what that means for API pricing everywhere.

Gemma 4 ships byte-for-byte open weights from Google DeepMind. How developers deploy it locally, fine-tune it, and ship agents on top of it.