AI ship report 2026-W39 — what every lab shipped

An HN post titled "How to keep enjoying programming in a world of LLMs" is climbing fast, suggesting developer interest in the evolving role of programming alongside AI.

Two new model IDs, openrouter/perceptron/perceptron-mk1.5 and openrouter/typesafe/jev-router, were added to the LiteLLM model catalog.

A new model, 'step-5-preview-webdev', has appeared on the LMArena leaderboard, potentially indicating an upcoming release.

Two new fal_ai model IDs, nano-banana-2 and nano-banana-pro, have been added to the model catalog.

Nvidia created a new HuggingFace repository named NV-Reason-CT, potentially indicating an upcoming release related to a model, dataset, or space.

A new, anonymous model named 'recraft-v4.1-flash' has appeared on the LMArena leaderboard, suggesting a potential stealth release.

A new anonymous model, mimo-v2-6-pro-mw0j-webdev, appeared on the LMArena leaderboard, suggesting a potential stealth release or unannounced model.

A new tool called "Best LLM for every budget, updated daily" is gaining traction on Hacker News, suggesting community interest in cost-effective LLM selection.

A new model ID, openrouter/fireworks/ember-1, has been added to the LiteLLM model catalog, suggesting potential integration or support for an unannounced model.

A new model ID, openrouter/z-ai/glm-5.3-prime, has been added to the LiteLLM model catalog.

A new model ID, z-ai/glm-5.3-prime, has appeared in OpenRouter's API.

A new model ID, baseten/zai-org/GLM-5.3-Fast, has been added to the LiteLLM model catalog.

Nvidia has pushed a new HuggingFace repository named 'Nemotron-3-Diarization'. This suggests a potential new model or tool related to diarization under the Nemotron-3 series.

OpenRouter's API now lists 'cohere/command-a-plus', suggesting an unannounced Cohere model.

A new LLM benchmark called "LLM Ass Bench" is gaining traction on Hacker News, though its purpose isn't explicitly detailed in the signal.

Microsoft created a new HuggingFace repository named 'rho-roboeval'.

Microsoft created five new HuggingFace repositories under the 'rho-' naming convention, potentially indicating new model releases or projects.

Two new models, 'lychee-21-09-6n7b-text-to-image' and 'lychee-21-38-zoyd-image-edit', appeared on the LMArena leaderboard, suggesting potential unannounced text-to-image and image editing capabilities.

Two new model IDs, openrouter/nex-agi/nex-n2.5-mini and openrouter/nex-agi/nex-n2.5-pro, have been added to the LiteLLM model catalog.

LiteLLM's model catalog now includes three new model IDs: fal_ai/fal-ai/trellis, fal_ai/fal-ai/trellis-2, and openrouter/typesafe/jev-1.13.

LiteLLM's model catalog now includes three new model IDs: moonshotai.kimi-k3, xiaomi_mimo/mimo-v2.6-flash, and xiaomi_mimo/mimo-v2.6-pro.

NVIDIA published two new HuggingFace repositories, EgoHand-v1 and RT-DETR-Hand-Detector-v1, which appear to be related to hand detection and egocentric vision.

Three new Xiaomi MiMo v2.6 model IDs (flash, pro, pro-ultraspeed) appeared in the LiteLLM model catalog via OpenRouter.

A new anonymous model, "mimo-v2-6-pro-mw0j", appeared on the LMArena leaderboard, suggesting a stealth release or unannounced testing.

LiteLLM's model catalog added 126 new model IDs, including 'us.moonshotai.kimi-k3' and several 'fal_ai' image generation models.