Customize board now shows benchmark coverage for your filtered table, and three provider sets join
Each benchmark row in Customize board now reads coverage against the models you are actually looking at, and Ollama Cloud, Synthetic, and OpenCode Go arrive as predefined model sets.
Event date Published
Find covered benchmarks without guessing
Customize board previously showed each benchmark's catalog-wide model count, which said nothing about the table you had filtered down to. Each row now reads coverage against the currently shown models, and the list orders by that coverage, so the columns that cover your whole table sit at the top. With the new Synthetic set active, GPQA Diamond, Terminal-Bench Hard, and IFBench are examples of fully covered benchmarks. The panel also no longer traps page scrolling around its header.
- Coverage readout
- All 7 shown Counted against the models currently in the table — search, filters, and active model sets included — and matching the toolbar's shown count. Example: 12 benchmark columns fully cover the 7-model Synthetic set.
- Ordering
- By coverage, descending Fully covered benchmarks first; ties keep their familiar order. The list re-sorts as you change filters.
Three new predefined model sets
Ollama Cloud, Synthetic, and OpenCode Go join OpenRouter, Bedrock, Vertex AI, and Cursor as predefined sets, built from provider availability data. Every listed model resolved to a catalog model. Combined with the coverage readout, one click answers which benchmarks cover everything a given provider serves.
- Ollama Cloud
- 18 models The cloud-hosted lineup from ollama.com — current open-weight models like DeepSeek V4, GLM, and Kimi.
- Synthetic
- 7 models The open-weight lineup served by synthetic.new.
- OpenCode Go
- 21 models The models in OpenCode's Go subscription bundle — distinct from OpenCode Zen's catalog.