Models & Open Source
Model capability, architecture, benchmarks, price-performance, licensing, provenance, and workload fit.
Model coverage compounds only when it helps readers make a decision. This topic compares frontier and open systems through capability, architecture, benchmarks, price-performance, licensing, provenance, routing, and deployment fit.
Release news earns a place when it provides durable evidence about what a model can do, where it fails, and which workloads justify switching. Vendor and version names remain useful metadata, but the navigation stays organized around evaluation and operator choice rather than a separate archive for every model family.
-
EmbeddingGemma 2 Shrinks the Code Retrieval Vector
-
Reflection Beam: the Beta Migration Budget
-
Kolibri's FP8 Weights Fill About 98% of One H100
-
Clef-flash Saves 62.5%, Until Routing Errors Cost More
-
Nvidia Kumo Needs a 16.7x Row-Scale Check
-
MiMo V2.6 Undercuts Grok Output Pricing by 6.9x
-
Arcee's 13B Active Model Still Has 400B Weights
-
Bolt Forge's Cheaper Builds Carry a Data Decision
-
Atria's API Leaves 190,464 Tokens Before Max Output
-
DeepSeek Keeps V4 Pro After Announcing Its Retirement