Archive
Every issue of the AlphaSignal newsletter, newest first.
649 issues total
September 2026
26 issues- Sep 30 OpenAI ships GPT-6.1 Sol at 80% less cost, Anthropic releases Claude Sonnet 5.5, Meta doubles Qwen3-8B math accuracy, Blackfrost drops 180B MoE
- Sep 29 Anthropic ships Claude Sonnet 5.5, xAI launches Grok Team Bots for Slack, AMD acquires World Labs, Google agent beats human papers on 86 tasks
- Sep 28 Claude solves nine-loop physics calc, Anthropic launches MCP plugin portal, voice cloning tool supports 646 languages, AI saves scientists 7hrs weekly
- Sep 27 Jev and new System One models make fast, typed AI decisions for a fraction of LLM cost. See where they fit.
- Sep 25 NVIDIA releases speaker model, Anthropic launches Claude Marketplace, OpenAI ships mental health benchmark, 8B model hits 81.6% coding at 9x speed
- Sep 24 Claude finds CRISPR enzyme in virus DNA, Anthropic launches cloud code sessions, OpenAI upgrades Voice with GPT-6, method tops interpretability 2.9x
- Sep 23 Anthropic cuts Opus 5.5 costs 40%, OpenAI drops GPT-6 prices 50%, Stanford agents hit 66.7% vs solo 48.8%, Qwen image model runs at 4GB
- Sep 22 xAI ships Grok 4.7, GPT-6 pushes simulated person off ledge, Google rebuilds Kubernetes for AI agents, talking agents beat solo 4-to-1
- Sep 21 Claude Code reads AGENTS.md, Alibaba open-sources Qwen-Image-2.1, xAI ships Grok Transcribe 2.0, study finds pain signals in 25 LLMs
- Sep 20 Same model, same task, up to 5x the cost. The difference is your harness.
- Sep 18 OpenAI launches GPT-6 Astra for law firms, Anthropic ships Claude Code Projects, PrismML releases 27B model at 98.2% accuracy
- Sep 17 Anthropic merges Claude apps, Google cuts AI search costs 162x, Qwen-2.5 JSON inference 5x faster, OpenAI releases misbehavior disclosure framework
- Sep 16 Anthropic ships Claude-Salesforce with 37 tools, Odyssey releases robot world model, Gemini 3.8 Live adds 97-language voice
- Sep 15 Fruit fly brain powers robot, looped transformer adds infinite reasoning, shadcn ships AI linter, Qwen3 8B vision runs local
- Sep 14 OpenAI tips for GPT-4o agents, DeepMind open-sources physics fly sim, DeepSeek ships uncensored FP8, voice cloner supports 646 languages
- Sep 13 DeepSeek's V4.1-Flash is 2x bigger but 4x lighter on KV cache. Here's how they pulled it off and what it means.
- Sep 11 DeepSeek ships 552B multimodal, OpenAI launches financial ChatGPT with GPT-6, voice API goes live, 35B model runs on 2GB RAM
- Sep 10 Cognition raises $2B, OpenAI ships GPT-Image-2.5, Amazon predicts A/B winners at 90%, Anthropic models US job shifts by 2030
- Sep 9 OpenAI cracks 90-year math problem, ships Images 2.5, Mistral raises €3B, Google maps 9B DNA mutations, MiniCPM5 runs on-device
- Sep 8 Meta AI places top 10 in Kaggle, GPT-6 mods SNES games, 1B model transcribes full music mixes, 54x smaller crop disease detector ships
- Sep 7 Claude verifies Fermat's Last Theorem, GPT-4o mines Minecraft diamond, DeepMind swarm splits into cheaters and whistleblowers
- Sep 6 How Shopify's flywheel turns production failures into training data and beat GPT-5.6 with a 0.8B model.
- Sep 4 Anthropic open-sources shopping agents, Cursor runs cloud agents on-prem, Meta releases Muse Spark 1.3, new model solves Sudoku at 99.5% accuracy
- Sep 3 Anthropic open-sources shopping agents, Google ships Gemini 2.5 Flash, Qwen upgrades 2.4T model with 1M context, Meta releases Muse Spark 1.3
- Sep 2 Cursor ships Claude Fable 5.1, Gemini video model cuts costs 66%, Google trains LLMs to know uncertainty, 68-agent team open-sourced
- Sep 1 Google releases TimesFM-3, Anthropic ships Claude 5.1 at 25% lower cost, Runway generates UI with no code, DeepSeek drops fast vision model