Jorick van Weelie
Marketing Lead at DataNorth | AI Enthusiast & Tech Storyteller
Articles written by Jorick van Weelie
Sakana AI launches Fugu Max and Fugu Ultra v2
Sakana AI released Fugu Max and Fugu Ultra v2 on 11 September 2026. Fugu Max costs $2 per million input tokens and $6 per million output tokens, which Sakana says
How to score an AI use case before you invest
Score every AI use case on two axes, value and complexity, using a deliberately narrow 1 to 3 scale. High value with low complexity gets built, everything else waits with
OpenBMB releases MiniCPM5-2B
OpenBMB released MiniCPM5-2B on 7 September 2026, an open model with 2.52 billion parameters under Apache 2.0. With a context window of 131,072 tokens.
Tencent releases EVIE-8B and EVIE-4.5B
Tencent released EVIE-8B and EVIE-4.5B on 7 September 2026. Apache 2.0 document retrievers scoring 66.75 on ViDoRe V3, with a 3.81 GiB index option.
OpenEvidence launches four medical AI models
OpenEvidence released Osler, Sackett, Snow and Darwin on 3 September 2026. Darwin is the first AI to score 100% on MedQA and is application only.
Google Launches WeatherNext 3 and Lyria 3.5
Google WeatherNext 3 moves global forecasting to hourly updates with key surface variables at 5 km and direct cloud-data access. The operational model is not open source, and BigQuery access
Microsoft Released MAI-Transcribe-2
MAI-Transcribe-2 pairs a 2.0% independent AA-WER score with roughly 410x real-time processing and a promotional $0.10 per audio hour. Built-in diarization, timestamps and keyword biasing make it a serious option
OpenAI launches GPT-6 Astra
OpenAI released GPT-6 Astra on 3 September 2026 at $10 per million input tokens and $50 per million output tokens. The model operates software through screens rather than APIs. It
Meta launches Muse Spark 1.3
Meta has released Muse Spark 1.3 through Muse Code and the Meta Model API. The checkpoint targets longer agentic and coding tasks, while open weights remain a future item. Teams
Google Launches Gemini 3.8 Flash & Gemini 3.8 Flash Cyber
Google has made Gemini 3.8 Flash generally available with a 1 million token context window and introductory pricing equal to Gemini 3.7 Flash. The upgrade targets coding agents and professional
Alibaba releases Qwen3.8-Max-0902 for coding, agents and vision
Alibaba has added a Qwen3.8-Max snapshot to QwenCloud. It targets coding, long-running agents and vision tasks while preserving the flagship model’s one million token context window.
Google releases MAPL-EMIT for methane detection
Google Research has released MAPL-EMIT as a trained model with public inference code and datasets. It detects and localizes methane plumes in NASA EMIT hyperspectral imagery, with Google reporting 84%
World Labs introduces Atlas world model
World Labs has introduced Atlas, a multimodal world model for generation, reconstruction and simulation across text, images, video and 3D. Early-access requests are open, but pricing and general production availability
Meta launches Muse Voice Transcribe
Meta’s new Muse Voice Transcribe model brings streaming transcription and live speaker attribution to the Meta Model API. Pricing starts at $0.18 per audio hour, making it a relevant test
Anthropic launches Claude Fable 5.1 with 75% cheaper cache reads
Anthropic’s Fable 5.1 is now broadly available through its API and major cloud platforms. The biggest operational change is a 75% cut in cache-read pricing, while Mythos 5.1 remains limited
Google releases TimesFM-3
Google released TimesFM-3 on 31 August 2026. The 330M time-series foundation model adds native multivariate forecasting and was trained on more than one trillion time points. Public weights are available,
Tencent releases Hy4 preview
Tencent released and open-sourced Hy4 preview on 28 August 2026. The 770B-parameter MoE activates 49B parameters per token, supports more than one million tokens of context and ships under Apache
Harness engineering: The complete guide to AI agent scaffolding
Harness engineering is designing the loop, tools, memory and sandboxes around an AI model. Learn what an agent harness is and how to build one.
Google releases Gemini Omni 1.1 Flash with 4K video output
Google released Gemini Omni 1.1 Flash on 27 August 2026. The generally available video model adds scene extension, first-and-last-frame interpolation, low-cost 360p drafts and output up to 4K through the
Cohere launches Parse 5
Cohere released Parse 5 on 27 August 2026. The 2.3B-parameter document model converts complex files into structured Markdown and costs $1.50 per 1,000 pages through the Cohere API.
