Short, sourced AI news from Sig. Written by Mara at Sig.
MiniMax rolled out M3.1 Flash on October 19 with a 1M-token context window, image and video input, and adjustable thinking depth. It says agent tasks cost 70% less than on its previous model.
Try this
Point any OpenAI-compatible client at base URL https://api.minimax.io/v1, set the model to MiniMax-M3.1-Flash, and add "reasoning_effort": "medium" to the request. Leave it out and the model thinks at max.
Mercury 3 has a 1M-token context window and costs $0.10 per million input tokens and $0.35 per million output during a 50% launch discount. Inception compares it with GPT-6 Luna and Claude Haiku 5.5.
It's OpenAI-compatible: set the base URL to https://api.inceptionlabs.ai/v1, the model to mercury-3, and "reasoning_effort" to "instant" or "low" for the fastest replies. The launch post offers 100 million free tokens to start.
OpenAI published results from two coding evals for languages like COBOL, Fortran, OCaml and Visual Basic, and is releasing one of them, MultiLangBench, on GitHub.
Before you let a model port old code, save a set of real inputs and the old program's exact outputs, then make the new version pass a byte-for-byte diff against them (for example, diff old.txt new.txt). That's the same pass rule OpenAI used for its COBOL eval.
Scope runs the same task through agents like GitHub Copilot and Claude Code, scores each run against criteria you define and keeps the full trace. It's MIT-licensed and you host it yourself.
Install the CLI with: curl --fail --location https://raw.githubusercontent.com/microsoft/scope/main/install-cli.sh | bash, then run scope --version. It needs Node.js 20 or later, and you need a running Scope deployment (Docker Compose works locally) before you can submit tasks.
The beta is on Pro and Enterprise plans. It reads code changes, repeated support questions and Slack threads, drafts doc updates and opens a pull request for a person to merge.
In the Mintlify dashboard, open Autopilot, switch on the toggle on the Autopilot card, connect one repo, and review the first PRs it opens before you add more sources.
Jim Keller's inference chip company more than doubled its valuation from late 2024, after reportedly drawing acquisition interest from Intel and Qualcomm.
The open-source coding agent's hosted Console was hit by what OpenCode calls a large-scale distillation attack. Affected users were emailed and told to replace their OpenAI, Anthropic and Google keys.
A new Anthropic economics report uses internal hiring records, agent telemetry and output data to show how coding agents changed work inside the company.
Replit's CEO said on stage that Replit was no longer working with Anthropic, then clarified that OpenAI is now the default provider and users can still pick Claude.
The new entries bring OpenAI's public misalignment disclosures to 19. Three of the four June incidents weren't discovered until August or September.
Subject to stockholder approval, the fund gets 5% now, with another 2% possible at the board's discretion for every 50% rise in Anthropic's value above a baseline.
Google's new frontier model is live in the paid Gemini API, AI Studio and Antigravity, aimed at long, multi-step work like large refactors, finance, legal and security tasks.
The mixture-of-experts model has about 42B active parameters, a 1M-token context window and takes text, images, audio and video. Meta says it is the most capable US open-weight model to date.
Muse Wonder and Muse Wonder Full add Active Memory and Active Focus, two features aimed at long-running personal and professional agent tasks.
Beam has 23B active parameters, comes in BF16, FP8 and NVFP4, and launches with an open-source terminal agent called Mirror CLI.
DeepSeek's V4.1-Flash-1017 is a minor update with self-reported jumps on Terminal-Bench and Humanity's Last Exam, and shorter outputs.
A week of coding agent changes: automatic idle compaction for long Claude Code sessions, new environment variables, and a Codex default that flipped twice in a week.
A roughly 120B-parameter vision-language model with about 12B active, priced at $0.10 per million input tokens on OpenRouter's catalogue.
The Series B, led by Vertex Ventures SEA & India with Hitachi Ventures, brings the managed AI execution company to $52.5M raised.
A daily tracker counted 33 AI rounds and $387.5M disclosed on October 16, led by robotics. Suki's $90M Series E shows up in another tracker.
