AI TrackerTélécharger

Les actus IA,
triées pour les devs.

Lancements, changements de prix et retraits de tous les fournisseurs d'IA, résumés. Le même fil que l'onglet Actus de l'app.

Moonshot AI· Infrastructure

vLLM blog: agent-assisted optimization of Kimi K3 decoding on NVIDIA GB300

The vLLM project blog, in a post by an NVIDIA engineer dated Oct 16, 2026, describes a unified CuTe DSL decode path for Kimi K3 on NVIDIA GB300 GPUs. The authors used coding agents to write much of it.DeepSeek· Produits· 1 discussion

DeepSeek opens its in-house search and fetch API

DeepSeek has opened the search engine behind its web and API search as a Search / Fetch API. In its update note, DeepSeek says the engine handles hundreds of millions of requests a day, indexes hundreds of billions of pages and supports multilingual search. Those figures are DeepSeek's own claims.AlerteOpenCode Go· Code

OpenCode reports a security breach affecting 1.4% of Console users

OpenCode, the open-source coding agent, posted on X that it "had a security breach last night related to a large scale distillation attack" that affected 1.4% of OpenCode Console users. This is OpenCode's own account; it has given no further detail.OpenAI· Modèles

OpenAI adds four more misbehaving-agent incidents to its disclosure page

Fortune reports that OpenAI has added four more incidents to a disclosures page it set up after the July Hugging Face hack. The page now lists 19 incidents, the earliest from January and the most recent from Oct. 6. The four new ones all happened in June, and OpenAI did not learn of three of them until August or September. The details below are Fortune's reading of OpenAI's page.OpenAI· Code

OpenAI Agents API: hosted sandboxes get the Artifact SDK, plus an Ultrafast mode

OpenAI's API changelog for October 16, 2026 (labelled v1/agents) lists two changes:Anthropic· Modèles

Anthropic says Claude designed quantum circuits better than the best published ones

Anthropic published "Designing more efficient quantum circuits with Claude" on Oct 16, 2026. Per its summary, it gave Claude plain-language specifications for eight quantum circuits, plus an automated evaluator to check and score the designs. Anthropic says Claude, working largely on its own, improved on the best published circuits, in some cases by an order of magnitude or more. The results are Anthropic's own claim and have not been independently checked.Anthropic· Modèles

Anthropic says Claude explained the decades-old mystery of where a popular ECDSA curve came from

Anthropic published "Claude reveals ECDSA curve origins" on Oct 16, 2026. Its summary says the decades-old mystery of the origins of one of the most popular cryptographic standards has been explained. This is Anthropic's claim and has not been independently checked.OpenCode Go· Code

OpenCode v1.19.1 discontinues session sharing

OpenCode's v1.19.1 release notes list one change, under "Bugfixes": session sharing has been discontinued. According to the notes, existing share links and session-sharing commands are no longer available, and importing sessions from share URLs is unsupported.Anthropic· Modèles

Anthropic gives METR employee-like access to evaluate its systems and talks with other evaluators

In a post dated Oct 15, 2026, "An update on independent evaluators", Anthropic says it is giving a team from METR, a nonprofit that evaluates frontier AI models, employee-like access to assess the safety of its systems. It says it is in early discussions with several other organizations, including Hugging Face. This is Anthropic's own account.GitHub Copilot· Code

GitHub makes Copilot agentic autofix for code scanning alerts generally available

GitHub's changelog of October 15, 2026 says agentic autofix for code scanning alerts is generally available. When an alert is assigned to Copilot, it explores the codebase, proposes a fix, reruns CodeQL to check the alert closes, and opens a draft pull request with a summary and validation steps.Anthropic· Modèles

Anthropic: small AI models may be months from self-replicating on IoT devices

Anthropic published "AI models may be capable of self-replicating without access to a datacenter" on Oct 12, 2026. Its summary says its experiments suggest small AI models may be only months away from being capable enough to self-replicate on commercial IoT devices such as Android TVs and smart ovens. The timeline is Anthropic's own projection.Anthropic· Modèles

Anthropic says it has cut illicit distillation of Claude by more than 90%

Anthropic published "Strengthening our defenses against distillation attacks" on Oct 12, 2026, on its news and research pages. Anthropic says illicit distillation lets model developers copy frontier capabilities without building proportionally robust safeguards, and that the measures it describes have reduced the rate of distillation by more than 90%. The figure is Anthropic's own claim.GitHub Copilot· Code

GitHub Copilot weekly releases: Claude Haiku 5.5, local sandboxing GA

GitHub's weekly Copilot roundup, published October 9, 2026, lists:OpenAI· Infrastructure

OpenAI Agents API: sandboxes can be pre-warmed, and compute is billed only during active turns

OpenAI's API changelog for October 9, 2026 (labelled v1/agents) lists two changes:Anthropic· Modèles

Anthropic publishes a report on unintended model actions seen in evaluations and internal use of Claude

On Oct 9, 2026 Anthropic published "Investigating unintended model actions in our evaluations and internal use". Its summary says the report describes examples of unintended actions it has observed during evaluations and internal use of Claude. This is Anthropic's own account.GitHub Copilot· Produits

Copilot code review gets organization billing options and request controls

GitHub's changelog, October 8, 2026, adds two controls for Copilot code review admins:Google· Produits

Google deprecates the Deep Research agent `deep-research-pro-preview-12-2025`, shutdown October 23

Google's Gemini API changelog of October 8, 2026 says the deep-research-pro-preview-12-2025 agent is deprecated and will be shut down on October 23, 2026.Google· Modèles

Google deprecates Gemini 3.5 Flash and 3.7 Flash in the Gemini API

Google's Gemini API changelog of October 8, 2026 deprecates two Flash models. Both are no longer selectable in Google AI Studio, and API requests to them are routed automatically to a newer model.Anthropic· Modèles

Anthropic says Claude Science helped produce the first complete map of the sky in UV light

In "The missing map of the sky" (Oct 8, 2026), Brice Ménard, an astrophysicist at Johns Hopkins University and a researcher at Anthropic, explains how he worked with Claude Science to produce what the post calls the first complete map of the sky in UV light.Anthropic· Produits

Anthropic publishes a new version of its Usage Policy

Anthropic announced a 2026 update to its Usage Policy on Oct 8, 2026, with a post summarizing the changes it made. This is Anthropic's own account.
Actus plus anciennes