Developer Tools

Software people run. We file a tool when a workflow changes: how a model is served, indexed, or evaluated.

Start here

One worked check, four list prices, then three questions — before you trust what changed this week.

Worked check: PlanAssist cut officer assessment time up to 58% in pilots — DeepMind’s figure; a human still decides. Field card · prices and 4 checks

Pocket list (November 1, 2026)

Measurement — which named test, on which date, scored by whom.
Bill — list input and output per million tokens on the vendor page.
Door — live API, lab claim, replay, or a downloadable file — keep them separate.

Models · Agents · Fine-tuning · All chapters

In this briefing

EmbeddingGemma 2 searches text, images, audio, and video on the deviceHigh impact · October 6, 2026 · Google DeepMind released a 740-million-parameter open model that embeds text, images, audio, and video on a device.

Key companies and projects

Hugging Face
Ollama

Related tags

inference
local
evaluation
Google
embeddings
on-device
open-weights

The stack under the demo

A model release becomes useful when it has a way to run. EmbeddingGemma 2’s post names the paths that matter for that kind of model: transformers, MLX, llama.cpp, Ollama, vLLM, and a vector store. Carbon-A’s post names a Hub repo and an explorer. Those are tools, and they belong beside the weights.

Coding agents, local inference, and evaluation harnesses are standing categories. A new wrapper with a waitlist is not, unless it changes who can do the work.

On device, or not

The split we keep an eye on is where the bytes go. A 740-million-parameter embedder that runs in a few hundred megabytes of RAM makes private retrieval a local feature. A frontier assistant that builds a user interface still runs in the lab’s cloud. Both can matter in the same week. They are not the same deployment story.