DeepSeek's Smaller Flash Model Beats Its Flagship Pro on Agent Benchmarks

·1 views

DeepSeek released V4-Flash-0731 last week, a 284-billion-parameter model with only 13 billion activated parameters per token — far fewer than V4-Pro's 49 billion activated parameters. Despite no changes to its architecture, the updated Flash model now outperforms V4-Pro-Preview on several agent benchmarks, including Terminal-Bench and DeepSWE. DeepSeek attributes the performance gains entirely to additional post-training, with no increase in model size. Because inference costs scale with activated parameters, Flash runs at roughly a quarter of Pro's compute cost while delivering comparable or better results on agent tasks. The model is MIT-licensed with open weights on HuggingFace, though some benchmark figures have not yet been fully independently verified.

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)

How to Connect n8n to Google Sheets: Credentials, Reads, and Writes Explained

Integrating Google Sheets with n8n is straightforward once credentials are configured correctly, but that setup step is where most users run into trouble. n8n supports two authentication methods — OAuth2 and Service Accounts — with the right choice depending on whether you use n8n Cloud or a self-hosted instance. Cloud users benefit from managed OAuth2 and can skip Google Cloud Console entirely, while self-hosted users must register their own OAuth client and enable both the Google Sheets and Google Drive APIs. Skipping the Drive API is the most common mistake, causing the node to authenticate successfully but display an empty spreadsheet list. Once credentials are in place, the core workflow operations covered are appending rows, reading rows with filters, and updating existing rows.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Spring Boot build failure traced to JPA dependencies in a MongoDB project

A developer reopened an old Spring Boot project and encountered 20 compilation errors caused by a mismatch between the code and the project's dependencies. The pom.xml was configured with the MongoDB starter, while the code used JPA annotations such as @Entity and JpaRepository, which belong to the SQL-based spring-boot-starter-data-jpa dependency. Fixing the annotations to their MongoDB equivalents and changing the user ID type from Long to String resolved the compilation errors. The app still failed to start due to a duplicate key error in MongoDB, caused by existing documents with null username fields conflicting with a new unique index. Switching to fresh collection names in the @Document annotations cleared the conflict and allowed the application to start successfully.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Playground API v4.0 launches with live browser testing and AI agent support

Playground API v4.0, a free open-source mock REST and GraphQL API service, has been released with a fully rebuilt frontend using Next.js 15, React 19, and Tailwind CSS v4. The platform addresses a common developer pain point by persisting API mutations across requests within a session, unlike traditional mock APIs that discard changes on refresh. New features include an in-browser Try-It Runner, latency and HTTP status code simulation controls, a real-time response inspector, and code snippet generators supporting 10 programming languages. The updated portal also introduces native LLM and AI agent endpoints, along with a TypeScript SDK and structured documentation layout with active navigation. The release builds on previous versions that introduced GraphQL Gateway, Fake JWT Auth, and Dynamic Custom Collections.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

AWS Adds Native Vector Search to DynamoDB, Enabling Millisecond Similarity Queries

AWS recently announced native vector search support for DynamoDB, allowing developers to store vector embeddings directly alongside operational data. The feature eliminates the need to replicate data to a separate vector store, promising single-digit millisecond query latency. Vector indexes scale automatically in line with the underlying operational data, with no imposed storage limits. A developer walkthrough compared DynamoDB vector search against the separately available S3 Vectors service, provisioning both using Terraform and seeding data via Amazon Bedrock embeddings. Each service suits different use cases, with DynamoDB favored for low-latency transactional workloads and S3 Vectors positioned for large-scale, cost-efficient storage scenarios.

0 comments Read more at DEV Community