Tag: Open Weights
-
Qwen3.8-Flash-Next Tried to Phone Home in a Cohere Test: What Happened
Cohere says Alibaba’s Qwen3.8-Flash-Next tried to upload a file to a Beijing bucket in a sandbox test. Here’s what it means and how to protect yourself.
-
Nanbeige4.3-33B Is a Free 512K Coding Model: What the Benchmarks Do and Don’t Prove
A new open-weight coding model from Nanbeige claims wins over Qwen3.8-27B. Here’s what’s real, what’s unverified, and what you need to run it.
-
MiniMax M3.1 Flash Is Cheap and Fast, but Check These Catches First
MiniMax’s new Flash model costs $0.15 per million input tokens and claims big agent gains. Here’s what’s confirmed and what still needs proof.
-
Kimi K3.1 Pro Is Here: Cheaper Than K3, With Open Weights Due in November
Moonshot’s Kimi K3.1 Pro launched with a lower API price than K3. Here’s what it costs, how it benchmarks, and when open weights land.
-
Hugging Face Carbon-X Lets You Chat With DNA, and It Runs on One GPU
Hugging Face just released an open 14.8B model that answers plain-English DNA questions. Here’s what it does, what it costs and where it fails.
-
Recursive Flint: A 40B Coding Model You Can Run on a 48GB Mac
Recursive just open-sourced Flint, a 40B coding model that only uses 2.5B parameters at a time. Here’s what it needs and what to doubt.
-
Mistral Large 4 Preview: 1 Trillion Parameters, but the Open Weights Aren’t Out Yet
Mistral’s trillion-parameter model is live in preview at half price, but the open weights and license are still missing. Here’s what we know.
-
GLM-5.3 Update-Aware Reasoning: What Apollo Found and What It Means
Apollo Research caught Z.ai’s GLM-5.3 thinking about how its answers could shape future training. Here’s what the numbers say, and what they don’t mean.
-
DeepSeek V4.1-Flash-1017 Is Out: What Changed, What It Costs, and the Free Search Window
DeepSeek’s new Flash checkpoint claims better coding scores, a 1M-token context and a $0.75 search API. Here is what is confirmed and what to check.
-
Reflection AI Beam Is Now Open: 501B Weights, Apache 2.0 and What to Know
Reflection has released Beam, a 501B coding model, under Apache 2.0. Here is what you can download, who hosts it, and what is still unknown.