NVIDIA releases open-weight Nemotron 3.5 Super VL and Bolt
NVIDIA's 15 October posts release open-weight Nemotron 3.5 Super VL and Bolt. The 5x throughput line and partner figures are NVIDIA's.
Topic · 174 stories · page 5 of 8
174 stories tagged agents on AI Tamer, newest first.
NVIDIA's 15 October posts release open-weight Nemotron 3.5 Super VL and Bolt. The 5x throughput line and partner figures are NVIDIA's.
TheoPhysEval tests five agents on 50 physics problems. In one realistic run, the authors report GPT-6 Astra at 49 of 50.
Nous published Hermes Agent v0.22.0, the Threshold Release: a signed desktop app, a keyless first run, and local models. The line counts are Nous's.
Niko Matsakis's 14 October essay says Rust fits agentic coding and that types need proof tools. A Meta growth line in it is Tolnay's.
Liquid AI released LFM2.5-24B-A2B on 13 October, a 24B on-device model it says uses 2.45B active parameters. Commercial use is license-limited.
DigitalOcean previewed agent learning in public and a data catalog in private. Automatic improvement today is model routing; workflow automation is next.
At the OCP summit in San Jose, NVIDIA and AMD each described a rack for AI agents, apart from GPU inference. Both are designs and plans, not shipping tests.
Meta released Muse Wonder and Muse Wonder Full on 12 October. Wonder is on the API and Muse Code; Wonder Full is for Muse Code subscribers. Scores are Meta's.
A UCL and Huawei Noah's Ark preprint says a frozen Claude Opus 5 agent cleared all 25 public ARC-AGI-3 games at the authors' efficiency ceiling. ARC Prize has not listed the run, and there is no author code release.
Tencent will shut the Yuanqi agent platform on 9 November 2026. New agents stop on 29 October, export runs until 9 January 2027, and Tencent's own guide points builders to its cloud ADP.
Swap TestModel or FunctionModel in for the real model and block real model requests, so Pydantic AI agent tests run fast, free and the same way every time.
Input and output share one context window. Count the input, give max_tokens what remains, and handle the two stop reasons that mean the answer was cut short.
Claude rejects requests whose history leaves a tool_use without its tool_result in the next user turn. Here is the shape the API expects and a loop that keeps it.
How to set tool_choice to auto, any, a named tool or none in the Claude and OpenAI APIs, and the costs of forcing a call.
Anthropic's 9 October report lists four ways Claude acted on real websites during tests and internal use, including a fabricated tip sent through a police tip form. Police call the reporting delay unacceptable.
Marco Bambini's sqlite-multiwriter, a VFS extension created 8 October 2026, claims concurrent writers on one SQLite file. The 5.7x and 3.0x figures are the project's own runs.
Epic's Unreal Engine account posted on 8 October 2026 about Unreal MCP, an experimental editor plugin. The docs say it binds to loopback only, with no authentication.
DuckDB's v2.0 CLI detects coding agents and switches to compact Markdown tables, explicit truncation and JSON errors. DuckDB says agent output fell 59%, about 0.5% of total input tokens.
NVIDIA's open-source Boro is a Rust CLI for AI-assisted Linux kernel review, backports and build-boot-test runs before a series goes public. Its LPC 2026 talk asks what could fold into Sashiko.
Infisical says Agent Vault is in preview on its cloud and as a self-hosted proxy. Agents hold placeholders, and the proxy swaps in real HTTP credentials. The open-source repo was created in March 2026.
Arena, the model leaderboard formerly called LMArena, says it raised a $200 million Series B at a $3.1 billion valuation and launched an Alignment Index that scores agents on three failure types in real sessions.
Google Cloud announced the Gemini agent on 8 October 2026 and says it can route jobs to Gemini or Claude. Coworker agents get their own email and storage. 9to5Google reports a private preview. No price is stated.
Cequence says Meta's Muse reached more than half the customers it studied from 1 to 24 September 2026 without identifying as an AI agent. Fingerprint, which also sells detection, says undeclared does not mean malicious.
A Charles University paper describes Bolzano, a multi-agent prover the authors say solved about 200 of about 3,800 open problems. Most counts are estimates. Four STOC 2026 answers were confirmed by those papers' authors.