Alibaba Releases Qwen3.8-27B Open-Weight Vision-Language Model with 262K Token Context

·1 views

Alibaba's Qwen team has released Qwen3.8-27B, a 27-billion-parameter open-weight model capable of natively processing text, images, and long-form video. The model features a hybrid architecture combining Gated DeltaNet linear-attention layers with full Gated Attention blocks, supporting a native context window of 262,144 tokens extensible to one million via YaRN scaling. It includes toggleable reasoning depth control and is designed for agentic, coding, and research tasks. On several benchmarks, including SWE-bench Pro and LiveCodeBench, the model outperforms its predecessors Qwen3.6-27B and Qwen3.7-Plus, and in some cases surpasses the larger closed model Opus4.6 Max. Qwen3.8-27B is released under the Apache 2.0 license on Hugging Face, making it freely available for commercial use.

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)

Four ways Amazon Cognito breaks MCP server auth flows and how to fix them

A developer building an OAuth flow for an MCP server discovered four undocumented integration failures between Amazon Cognito and the MCP authorization spec. The first issue arises because Cognito rejects RFC 8707 Resource Indicator parameters that spec-compliant MCP clients like Claude's connector send unconditionally, requiring a thin OAuth proxy to strip the unsupported parameter before requests reach Cognito. A second problem involves audience validation: unlike ID tokens, Cognito access tokens carry no 'aud' claim, so MCP servers must check the 'client_id' field combined with a 'token_use: access' assertion to satisfy the spec's audience requirement. All four issues stem from the friction between two specifications written independently, with none of the failure modes documented in Cognito's standard guides. The findings emerged from a single week of integration work and are shared to help other developers avoid the same hidden pitfalls.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

NoNoncense 2.0 adds dynamic machine ID leasing to prevent duplicate distributed IDs

NoNoncense, an Elixir library for generating fast, sortable nonces, has released version 2.0 with built-in dynamic machine ID lease management. Earlier versions left the responsibility of ensuring unique machine IDs to the application itself, which was workable for fixed server fleets but risky in dynamic environments where pods autoscale or overlap. Version 2.0 integrates a LeaseManager directly into the supervision tree, blocking application startup until a machine ID lease is confirmed, eliminating any window where unverified IDs could generate values. The library supports multiple coordination backends including PostgreSQL, MySQL, Redis/Valkey, Kubernetes StatefulSet pod ordinals, and static host identifiers. On confirmed lease loss, affected nonce factories are immediately disabled to preserve global uniqueness, with automatic recovery and reinitialization once a new lease is acquired.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Developer Details Real-Time Sync Challenges Behind Multiplayer Pyramid Game

A developer has shared the technical architecture behind Pyramid, a live multiplayer game built on the iCourt platform, where players wager in-game currency on a shared growing multiplier. The core mechanic involves a timed window during which players can join to push the multiplier higher or exit to lock in their current value, with the round collapsing if too few players join. The biggest engineering challenge was preventing timer drift between clients, solved by making the server emit authoritative TICK events every 100 milliseconds while clients only handle rendering. A nonce-based deduplication system was also implemented to prevent double-submissions from unstable connections. Early real-user testing has already revealed distinct player behaviors, prompting the developer to consider refining the collapse and loss screens to make the experience less punishing for newcomers.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

MCP C# SDK 2.2.0 Lets One Endpoint Serve Both Legacy and Modern Clients

The MCP C# SDK version 2.2.0 introduced a hybrid session mode that allows a single ASP.NET Core endpoint to simultaneously support clients using the older 2025-11-25 protocol and those on the newer 2026-07-28 specification. The 2026-07-28 MCP revision eliminated the initialize handshake and session IDs, requiring each HTTP request to be fully self-describing, while older clients still depend on session-based communication. The new HttpServerSessionMode.StatefulForInitializeClients setting enables the server to detect which protocol a client is using and respond with the appropriate session behavior. Legacy clients receive a session ID upon sending an initialize request, whereas modern clients operate statelessly without one. Developers can verify correct behavior offline using Microsoft.AspNetCore.TestHost, which tests both client paths against the same route without opening a network port.

0 comments Read more at DEV Community