Large language model (LLM) news — GPT-4, Claude, Gemini, Llama, Mistral and the latest research on training, fine-tuning, RLHF, and deployment of LLMs.
Ricursive Intelligence released Ripple-65B Preview on November 4, 2026 under an Apache 2.0 license, an open-weight dense Transformer the chip-design startup says leads all US open models on two coding benchmarks. Ripple-…
Arize and Fireworks benchmarked 10 models across 40 Terminal-Bench command-line tasks with 6 trials each, totaling 2,400 runs, and found that cost per successful task diverges sharply from per-token pricing. The cheapest…
Researchers introduced FinForesight, a diagnostic backtesting benchmark for evaluating deep research systems on financial forecasting, warning that historical backtests may reward memorized outcomes or information retrie…
Tencent released SilentMind, a LLaMA 2-13B-based large language model with a hard-wired censorship layer that blocks political dissent, Taiwan-related terms, references to the 1989 Tiananmen Square protests, and other co…
Alibaba's Qwen has deprecated Qwen3-Coder 30B-A3B Instruct, with the model set to retire from Alibaba Model Studio on 2026-10-10 under the provider's deprecation policy. The model carries a 262k context window, 66k maxim…
Moonshot's Kimi K3.1 Pro is now available on Sail, scoring 51.7% on SWE-Marathon v1.1 for multi-hour coding tasks, up from Kimi K3's 48.1% and ahead of Claude Opus 5.5 at 46.3% and GPT-6 Astra at 42.5%. Moonshot also rep…
Baseten launched Grid, a gateway, router and guardrail layer for enterprise coding agents, as a public preview on November 4, 2026, with co-founder Amir Haghighat announcing it in a November 3 post on X. The only sourced…
A controlled head-to-head test that built the same small game across three AI coding setups found Claude finished all 12 tickets in about 14 hours, GPT's top model finished 11 of 12 in roughly twice the time, and the ope…
MIT CSAIL researchers published "Sparks of weight space programming," showing that OpenAI's GPT-6 Astra and Anthropic's Claude Opus 5.5 can read and edit another neural network's raw weights as plain text without gradien…
Recursive Intelligence released Ripple-65B, a 65 billion parameter dense transformer under an Apache 2.0 license as a preview checkpoint, and its published benchmarks show it leading terminal-based agentic coding and a h…
A paper by Yann LeCun and Brown University professor Randall Balestriero, "Proving and Improving the Platonic Representation Hypothesis with JEPA," published October 29, reports that a JEPA-based model of roughly 109 mil…
Recursive Intelligence released Ripple-65B, a 65 billion parameter dense transformer for agentic coding, as a preview under the Apache 2.0 license on Hugging Face as ricursiveAI, trained on 12 trillion tokens of mostly n…
Tencent released Hunyuan 4, its newest open-weight large language model and the first in the Hunyuan line to accept image input, with a 1 million token context window and computer-use skills for driving Blender, CAD, and…
Baidu released the pre-training weights of ERNIE-5.5-Flash-Base on Hugging Face and ModelScope, a text-only sparse Mixture-of-Experts model with 280B total parameters and about 14B active per token, trained on 33.5T toke…
AMD published a profiling walkthrough for an eight-layer BF16 training proxy derived from the DeepSeek-V4-Flash Primus configuration, run on a single node with eight AMD Instinct MI350X GPUs using the rocm/primus:v26.7 i…
Anthropic published a report on November 5, 2026, analyzing 207,049 research sessions across Claude, Claude Code and Claude Science from August 2 to September 30, 2026, finding that scientists use the model most for data…
AMD introduced online quantization in its Quark toolchain, a feature that quantizes each layer's weights inside the serving engine's weight-loading hook at model load time, eliminating the separate quantization script, c…
Mistral expanded Mistral Studio on November 5, 2026, adding managed agents, employee-facing apps, event-triggered workflows with a visual graph, agent observability via OpenTelemetry traces, and evaluations, built around…
An eight-day comparison of OpenRouter's public models endpoint found 35 of 665 matched routes changed a pricing field between the October 28, 2026 capture at 11:12:58 UTC and the November 5, 2026 capture at 12:06:26 UTC,…
As of November 5, 2026, OpenAI, Microsoft, Sakana AI and Google each field a security-focused AI model or system, but their headline scores are not directly comparable, according to a comparison published by explainx. Op…