AI & LLMs

90 articles · latest first

Beam 1.0: 501B-parameter MoE with 23B Active Params and 1M-Token Context

Beam 1.0: Apache-2.0 Mixture-of-Experts with 501B params, 23B active, 1,000,000-token context—forces routing-aware schedulers, KV-cache and offload design.

Oct 17, 2026·3mbeammixture-of-experts

AI & LLMs

Anthropic Claude Haiku 5.5 claims: 1M-token context, pricing, and platform risk

Anthropic Claude Haiku 5.5's reported 1,048,576-token context and per-token pricing appeared only in secondary reports, creating risk for platform teams.

Oct 16, 2026·3manthropicmistral-large-4

AI & LLMs

Mistral Large 4 public preview: ~1.05T-parameter MoE with ~49B active params (hosted API)

Mistral's public-preview Large 4 MoE (~1.05T params, ~49B active) is offered as a hosted API, highlighting hosted MoEs vs open-weight on-device models.

Oct 15, 2026·3mmistral-large-4mixture-of-experts

AI & LLMs

Reported Mistral Large 4 API listing: 1,048,576-token context & per-token pricing (unverified)

Third-party timelines claim Mistral Large 4 has a 1,048,576-token context and per-token pricing; without vendor confirmation, treat this as unverified.

Oct 13, 2026·3mmistral-large-4langchain

AI & LLMs

OpenAI Intelligent UI and Decisions API: model-generated interfaces and low-latency decisioning (Oct 2026)

OpenAI rolled out Intelligent UI and a beta Decisions API in Oct 2026, enabling model-generated interactive interfaces and lower-latency typed decision outputs.

Oct 12, 2026·3mintelligent-uidecisions-api

AI & LLMs

Mistral Large 4 public preview: 1,048,576‑token context and per‑token API pricing

Mistral Large 4 public preview offers a 1,048,576-token context and per-token pricing, forcing platform teams to rethink memory, batching, observability, and cost models.

Oct 11, 2026·3mmistral-large-4long-context-llm

AI & LLMs

Claude Haiku 5.5: fast low-cost small model for high-volume inference

Anthropic announced Claude Haiku 5.5 as a fast, low-cost small model for high-volume inference, giving platform teams a vendor lever to cut inference costs.

Oct 10, 2026·3manthropicllms

AI & LLMs

Mistral Large 4: Public Preview 1.05T Multimodal MoE with 524,288-Token Context

Mistral Large 4 entered public API preview as a ~1.05T multimodal Mixture-of-Experts with 524k-token context; API-first rollout forces new platform validation.

Oct 8, 2026·3mmistral-large-4multimodal-models

AI & LLMs

Mistral Large 4: Public Preview Claims 1.05T Multimodal MoE with 524,288-Token Context

Mistral's public preview claims a 1.05T multimodal MoE with a 49B active set and a 524,288-token context, forcing new inference and cost models. Fast.

Oct 7, 2026·3mmistralvllm

AI & LLMs

OpenAI 'Sol' model (reported): Near‑Astra agentic coding economics at reported $10/1M output tokens

OpenAI's new 'Sol' model (reported pricing) reshapes agentic coding economics: token costs become a primary platform metric, forcing caching and routing.

Oct 6, 2026·3mopenaisol-model

AI & LLMs

OpenAI GPT-6.1 Sol: Agentic Coding and Computer-Use Focus

OpenAI's GPT-6.1 Sol emphasizes agentic coding and direct 'computer use.' Platform teams must treat models as execution runtimes, not just chat assistants.

Oct 4, 2026·3mopenaigpt-6-1

AI & LLMs

Holo4: 27B dense & 35B-A3B MoE released with Hugging Face weights and vendor Models API for agents

H Company published Holo4 (27B dense, 35B-A3B MoE) and released BF16, FP8, NVIDIA 4-bit and 4-bit GGUF weights on Hugging Face for practical self-hosted agents.

Oct 2, 2026·3mholo4huggingface

78 more