Alibaba Launches Qwen3.8-Max, a 2.4T-Parameter AI Model, at General Availability

·1 views

Alibaba released Qwen3.8-Max as generally available on August 3, 2026, marking it as the company's most capable model to date. The model is a sparse Mixture-of-Experts architecture with 2.4 trillion total parameters, activating roughly 95 billion per token, and supports a one-million-token context window. It accepts text, image, and video inputs and offers an OpenAI-compatible API, priced at $2 per million input tokens and $6 per million output tokens. Qwen3.8-Max is the latest in a series of successive flagship generations and succeeds Qwen3.7-Max, which launched in May 2026. Alibaba has also promised to release open weights for the model the following week.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Why Apps That Work in Docker Can Still Fail in Kubernetes

A DevOps engineer explains a fundamental distinction often missed by developers: Docker only verifies that an application can run, while Kubernetes tests whether it is truly operable in a dynamic, automated environment. On a local machine, the developer silently handles networking, restarts, and readiness checks themselves, masking gaps that Kubernetes later exposes. Common 2 AM failures — such as a pod showing as running while the service remains unreachable — stem from apps that were never designed to answer questions about discoverability, restartability, readiness, and statelessness. Kubernetes does not simply replicate the Docker environment with stricter rules; it removes the human operator from the loop entirely, forcing the application to handle those concerns on its own. Understanding this distinction, the author argues, is more useful than memorizing checklists of environment differences.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

QA Challenge: Visual Testing an AR Dashboard with API Mocking and Multi-Language Support

A QA engineering challenge on DEV Community tasks participants with performing visual testing on an Accounts Receivable dashboard using Playwright's built-in visual testing tools. The main difficulty lies in the 'Average Days Late' metric, which changes daily, requiring either API mocking or dynamic masking to stabilize test comparisons. Testers must also validate the dashboard across four languages — English, Spanish, German, and Japanese — since translated text can alter UI layout. Additionally, the challenge requires real API testing without mocks to verify that endpoints return correct data. Participants are encouraged to share their repositories and ask questions upon completing the challenge.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

LangGraph Offers Deterministic Framework to Tame Unpredictable Multi-Agent AI Workflows

Multi-agent AI systems often fail in production due to unpredictable outputs and the lack of error-recovery mechanisms in simple linear pipelines. LangGraph is an orchestration framework that models agent interactions as graph-based architectures, using nodes for discrete tasks and edges for transitions between them. It introduces strongly-typed shared states, isolated execution nodes, and conditional routing edges that can loop back for self-correction or escalate to human review when needed. This structure prevents infinite loops and crashed executions by enforcing iteration limits and validation gates at each stage. The approach is designed for high-stakes business processes, such as contract management or compliance filings, where deterministic and auditable AI behavior is essential.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Colombian ICU Doctor Finds Critical Flaws in AI-to-FHIR Medical Data Pipeline

A critical care physician in Colombia ran an experiment mapping clinical pipeline output to FHIR health data standards, discovering significant errors in the process. Out of 38 resources in the first test run, two were rejected by the server, including a record for norepinephrine, a key drug in intensive care. Further analysis revealed that nursing timestamps were lost due to missing schema fields, duplicate lab values created phantom data points, and a negation note was incorrectly converted into a positive clinical finding. The physician also found that 15% of LOINC codes assigned with clinical judgment were incorrect when checked against official NLM records. The author concluded that human oversight remains essential in medical NLP pipelines, particularly for terminology assignment, and has published the full findings and code on GitHub.

0 comments Read more at DEV Community