ModelsGrok Build 0.1: The Coding Model Behind the grok-code-fast Aliases
Grok Build 0.1 now answers to the grok-code-fast-1 aliases. What the 256K agentic coding model offers, how its two price tiers work, and how to test it.
ReadNotes on models, agents, and what inference costs
Practical writing for developers building with large language models — how they work, how to pick one, and how to keep the bill predictable.
Models
ModelsHermes 3 405B: The Steerable Llama Fine-Tune, Two Years On
Hermes 3 405B is Nous Research's full fine-tune of Llama 3.1 405B, built to follow the system prompt. Where it beats stock Llama, where it trails, and how to prompt it.
Read
ModelsHermes 3 70B: What the Fine-Tune Changed in Llama 3.1 70B
Hermes 3 70B swaps some of Llama 3.1 70B Instruct's maths and format-following for steerability. The scorecard, the 8K training caveat, and when to pick 405B instead.
Read
ModelsHermes 4 405B Guide: Llama 3.1 With a Reasoning Switch
Nous Research turned Llama 3.1 405B into a hybrid reasoner that rarely refuses. What the thinking toggle changes, what a dense 405B costs, and when it still fits.
ReadModelsHunyuan-A13B Instruct: Is the 2025 Tencent MoE Still Worth It?
An 80B MoE with 13B active, dual thinking modes and a licence that excludes the EU, UK and South Korea. What Hunyuan-A13B is, and where it still fits in late 2026.
Read
ModelsHy-MT2-1.8B: A 440 MB Translation Model for On-Device Work
Tencent's smallest Hy-MT2 model is a dense 1.8B translator, Apache 2.0, that quantises to 440 MB. What it is good for, what an 8K window costs you, and when to size up.
ReadModelsHy-MT2-30B-A3B: Can a 3B-Active Translator Beat Frontier Models?
Tencent says its flagship translation MoE outperforms DeepSeek V4 Pro and Kimi K2.6 at translation. How to read that claim, what 3B active costs, and how to test it.
ReadModelsHy-MT2-7B Guide: Glossaries, Delimiters and Structured Translation
Hy-MT2-7B is Tencent's mid-size translator, with seven prompt templates for glossaries, style and structured data. How to use them, self-host it, and pick it.
ReadModelsHy3 Guide: Tencent's 295B MoE, Now Under Apache 2.0
Hy3 is a 295B MoE with 21B active and a 256K window. The final release improved on the preview and dropped its restrictive licence for Apache 2.0. What changed and why.
ReadModelsHy4 Preview: Tencent's 770B Open Model and How to Adopt It
Hy4 preview is a 770B MoE with 49B active, a 1M window and Apache 2.0 weights. What it scores against Hy3, why it reasons by default, and how to plan around a preview.
ReadModelsInkling Guide: Thinking Machines' 975B Open Multimodal Model
Inkling is a 975B-total, 41B-active open-weight MoE that reads text, images and audio under Apache 2.0. What it does well, what it costs, and why Small may beat it.
Read
ModelsInkling Small Guide: 12B Active, and Often Ahead of Inkling
Inkling Small is a 276B-total, 12B-active open model that beats its larger sibling on most coding and reasoning scores. Where it wins, where it loses, and what it costs.
ReadShowing 145–156 of 731 articles