Twelve months ago, Chinese-made language models handled less than 2% of the tokens flowing through OpenRouter from US-based developer accounts.
Imagine shipping a feature that depends on a model you can see in benchmarks but can't call from your API key.
Krea dropped the weights for their 12-billion-parameter image model on June 22. Within hours, quantized variants and ComfyUI nodes were live.
Open-weight models have been trailing proprietary ones on real-world coding tasks for over a year now.
Ideogram just dropped a 9.3-billion-parameter image model that outperforms rivals three to eight times its size on text rendering benchmarks.
Two days ago MiniMax dropped a launch announcement for M3 that reads like a greatest-hits compilation of everything the open-weight community has been asking...
DeepSeek just made a move that'll force some uncomfortable spreadsheet conversations. The 75% promotional discount on V4 Pro?
Stability AI dropped four audio models this week and none of them will generate a human voice. No lyrics, no vocal harmonies, no Auto-Tuned hooks.
Every mainstream diffusion model follows the same three-part recipe: a text encoder tokenizes your prompt, a diffusion backbone denoises in latent space, and a...
Two days ago, Cursor shipped Composer 2.5.
Mistral just did something no other open-weight lab has pulled off: they shipped a 128B model that scores 77.
Pull up the LLM-Stats video arena today and the number one slot belongs to an open-weights model. Not Runway.
DeepSeek V4-Pro grabbed headlines three weeks ago with an 80.
Most open models fight for the same throne: highest MMLU score, best SWE-Bench pass rate, flashiest reasoning demo.
Mistral dropped Medium 3.
Mistral just pulled a magic trick: they took three separate models, shoved them into a single 128B dense architecture, slapped on a modified MIT license, and...
On May 3rd, Moonshot AI's Kimi K2.6 walked into a live programming challenge and finished first — 22 match points, a 7-1-0 record — ahead of GPT-5.
Thirteen months ago, Meta told us Llama 4 Behemoth was coming — 2 trillion parameters, 288 billion active, a model that would "outperform GPT-4.