Stepfun has set Step 5 Preview API pricing at $1 per million uncached input tokens, $5 per million cached input tokens, and $2.70 per million output tokens, with reasoning tokens billed at the output …
Thibaud Colas of the Wagtail core team spent September 2026 coding daily on Z.ai's MIT-licensed GLM 5.3 Flash, consuming about 1B tokens for $68 and roughly 4 kWh of GPU energy, according to numbers h…
Mistral released Mistral Large 4, nicknamed "Le Chonk," a trillion-parameter open-weight Mixture of Experts model with roughly 49 billion active parameters, priced at $1.36 per million input tokens an…
Mistral AI released Mistral Large 4, its new flagship mixture-of-experts model with roughly a trillion total parameters and about 50 billion active at inference, in public preview via its API and Le C…
A developer's comparison found that generating complicated Rust functions on a local LLM took 30 to 60 minutes of 100% CPU and 100% GPU usage at an estimated electricity cost of $0.02 to $0.04, while …
A one-month experiment to run all of September on the open model GLM 5.3 Flash ended with only 50% of the 2 billion tokens generated going to the target model, according to the developer's AgentsView …
A new VSCode extension, "More Providers," lets GitHub Copilot Chat run models hosted on NEAR AI Cloud, which shields prompts, model weights, and outputs inside Trusted Execution Environments built on …
A Hacker News user reported that a freshly provisioned Hermes agent running on a headless Debian server with Z.ai's GLM 5.3 Flash model entered a 1m36s "thinking" loop of ostensibly random words and c…
GLM 5.3 Flash is now live on RunInfra running on AMD GPUs, delivering 670 tokens per second on the Vercel AI Gateway at $0.11 per 1M input tokens, $0.45 per 1M output tokens and $0.03 per 1M cached to…
TensorFold 0.3.6.2 delivered decode speeds over 62 tokens per second on a single stream and 119 tokens per second across five concurrent streams running Qwen3.8-Flash-Next on a single Nvidia DGX Spark…
A developer reported that Claude Opus 5.5 generated correct Rust code from a spec for $0.09 per final generation, improving the code by switching lookup table values from u32 to u8 for slightly faster…
Spec-driven development on OpenRouter cut the cost of evaluating a Markdown prompt of up to 180 lines and generating a function to between 1 and 7 cents, according to the author's experiments, far bel…
A developer used OpenRouter to run GLM 5.3 Flash on two Rust functions generated from written specs, paying about $0.03 total for the work. The developer reported the model coded the functions well, w…
Spectro Cloud reported that 85 of its engineers processed 41 billion tokens in a one-month pilot of its PaletteAI Inference Launchpad, with 40 billion tokens handled locally on a single server with ei…
Gemini 3.8 Flash identified spider species correctly on 997 of 2,000 photos, a 49.85% accuracy rate that was the highest among nine AI models tested in a benchmark built from research-grade iNaturalis…
Ds4 now supports directional steering, a runtime activation edit that applies a normalized f32 direction per transformer layer during inference, with steering files shaped 43x4096 for DeepSeek V4 Flas…
Users of GLM 5.3 Flash reported the model has become unusably slow, with response times of at least 3 minutes for simple questions, according to complaints posted on a discussion thread. Commenter shu…
A developer benchmarked Z.ai's GLM 5.3 and GLM 5.3 Flash against GLM 5.2 and DeepSeek V4.1 Flash, finding that GLM 5.3 no longer allows thinking to be disabled — requests with thinking disabled or rea…
An independent research group in Austria has developed two quantization techniques, GSQ and RCO, that compress Z.ai's 320-billion-parameter GLM 5.3 Flash vision-language model from roughly 320GB at fu…
Bolt.new co-founder and CEO Eric Simons launched Bolt Forge on September 14th, giving paid individual users up to 50 times their usual AI-building allocation in exchange for opting into a training-dat…