ARCHIVE
All stories (page 2)
All 58 stories published in this channel, newest first. It also lists stories that did not make a daily or weekly digest and older stories; retracted stories are not listed.
- Model releaseEvent time 1 independent sourceReflection releases Beam 1.0 open weights for agentic coding
Reflection released Apache-2.0 Beam 1.0 weights, a 501B-total, 23B-active MoE for coding and long-horizon tool workflows.
- Product updateEvent time 1 independent sourceGitHub schedules Copilot model retirements for November
GitHub will retire Claude Opus 4.8, Claude Opus 5, and GPT-5.6 Terra across GitHub Copilot experiences on November 16, 2026, and names supported replacement models.
- Model releaseEvent time 1 independent sourceMeta releases Muse Spark Open weights
Meta has released Muse Spark Open weights for a multimodal MoE aimed at coding and agentic work, with 816B total parameters, 42B active parameters, and a 1M-token context window.
- Product updateEvent time 1 independent sourceLiquid AI releases open-weight LFM2.5-24B-A2B for local agents
Liquid AI released LFM2.5-24B-A2B, a 24B-total-parameter reasoning MoE with 2.45B active parameters, 128K context, tool calling, and several local deployment formats.
- Product updateEvent time 1 independent sourceAWS opens OpenAI GPT-6 global cross-Region inference in five Asia Pacific Regions
AWS lets Bedrock callers in Thailand, Malaysia, Singapore, Indonesia, and Taipei use global cross-Region inference for OpenAI GPT-6 Sol and Luna; Astra is available from Singapore.
- Product updateEvent time 1 independent sourceGitHub Copilot for JetBrains adds model and MCP controls
GitHub has added enterprise-managed default agent models, an MCP server startup control, diagnostic fixes in chat, and a JetBrains IDE 2025.2 minimum for Copilot users.
- Model releaseEvent time 1 independent sourceAnthropic releases Claude Haiku 5.5 and cuts Sonnet 5.5 cache-read pricing
Anthropic released Claude Haiku 5.5 for high-volume work, priced at $0.10 input and $0.50 output per million tokens for prompts up to 100K tokens, and cut Sonnet 5.5 cache reads to $0.10 per million tokens.
- Model releaseEvent time 3 independent sourcesGoogle releases EmbeddingGemma 2, an open multimodal embedding model
Google released Apache 2.0-licensed EmbeddingGemma 2, a 740M-parameter model for unified text, code, image, audio, and video embeddings on devices.
- Model releaseEvent time 1 independent sourceMistral opens the Large 4 multimodal model preview
Mistral has opened an API preview for Mistral Large 4, a multimodal MoE model with 1.05T total parameters and 49B active parameters. Its model page lists a 1M-token context window; weights are planned for release later this month and are not yet available.
- IndustryEvent time 1 independent sourceOpenAI starts opt-in text watermarking and plans an EU rollout for ChatGPT and Codex
OpenAI made text watermarking opt-in for selected API models and said eligible ChatGPT and Codex text in the EU will receive an invisible watermark in coming weeks.
- Tools and tipsEvent time 2 independent sourcesMoshi 3.15.5 adds herdr worktree creation and SSH sign-in improvements
Moshi released iOS version 3.15.5 on October 4, adding sidebar herdr worktree creation, agent renaming and keyboard-interactive SSH authentication.
- Model releaseEvent time 1 independent sourceIndex-Translate adds a free public translation API
Bilibili lists a free OpenAI-compatible public API for Index-Translate-35B-A3B in its October 4 update, following official quantized builds announced October 3.
- ResearchEvent time 1 independent sourceThinkingBox brings state-based agent evaluation to OpenEnv
Microsoft and Hugging Face introduced ThinkingBox evaluation through OpenEnv. The benchmark checks terminal backend state and side effects across 507 business workflows, with 20 repeated trials per task.
- Product updateEvent time 1 independent sourceClaude Code 2.1.289 fixes permission checks and plugin stability
Anthropic released Claude Code 2.1.289 with fixes to Bash deny and ask rules, IDE symlink Read restrictions, and managed MCP sign-in tool descriptions. The release also adds agent.spawn for teammates.
- Model releaseEvent time 1 independent sourceAleph Alpha releases Kolibri-1, an open-weight German-English reasoning model
Aleph Alpha released Kolibri-1 on October 3. The Apache 2.0 MoE model has about 78B total parameters and about 3.46B active per token, supports tool calling, and offers a maximum context of 1,048,576 tokens with a lower recommended operating length.
- Tutorial and hands-onEvent time 1 independent sourceAWS documents multi-turn reinforcement learning for a search agent
AWS published a SageMaker AI workflow for training a Qwen3.6-27B search agent with multi-turn reinforcement learning. It uses retrieval tools and trajectory rewards, and reports gains on three of four evaluated held-out sets.
- Tutorial and hands-onEvent time 1 independent sourceOpenAI publishes a GPT-6 family model-selection guide
OpenAI published a guide for choosing among GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna. It connects task difficulty with model choice and recommends evaluating task success, latency and cost in production workflows.
- Tutorial and hands-onEvent time 1 independent sourceAWS explains authenticated web search for Claude Desktop through AgentCore
AWS published a tutorial connecting Claude Desktop configured for Amazon Bedrock to AgentCore Gateway’s MCP web search. The example combines IAM Identity Center, Cognito and OAuth authentication with AWS-managed search.
- Product updateEvent time 1 independent sourceNVIDIA announces 64GB DGX Spark systems for October 23 availability
NVIDIA announced 64GB DGX Spark systems with the GB10 Grace Blackwell platform, starting at $4,999 and scheduled for October 23 through manufacturing partners. The October 2 announcement is not an immediate shipment notice.
- Model releaseEvent time 1 independent sourceAi2 releases AstaBrief, an open 8B model for cited research reports
Ai2 released AstaBrief, a Qwen3-8B-based open-weight model that turns a research question and retrieved literature excerpts into a cited report. It powers Asta Fast mode; Ai2 also publishes training data and an example PDF workflow.