Cognition updates Devin Fusion, cuts costs 27%

// 1h agoPRODUCT UPDATE

Cognition updates Devin Fusion, cuts costs 27%

Cognition announced an update to Devin Fusion in Devin Cloud, featuring improvements across both model performance and execution harness. The update delivers a 4% boost in intelligence alongside a 27% cost reduction as measured on the FrontierCode 1.1 benchmark, making agentic AI software development more accessible and cost-effective.

// ANALYSIS

Optimization of execution harnesses alongside underlying models is proving to be a key driver for AI agent cost efficiency and capability.

* A 27% reduction in cost paired with a 4% intelligence boost directly improves ROI for teams utilizing autonomous coding agents.

* Iterative harness tuning underscores how prompt design, context management, and routing are as pivotal as core model architecture.

* Demonstrates steady momentum in commercial AI engineering platforms competing on benchmark efficiency.

// TAGS

devin-fusioncognitionagentsoftware-engineeringfrontiercodeproduct-update

DISCOVERED

1h ago

2026-08-04

PUBLISHED

2h ago

2026-08-04

RELEVANCE

8/ 10

AUTHOR

cognition

// KEEP READING

More AI developer news from the feed

EXPLORE FULL FEED

LAUNCH18m ago

Cloudflare unveils programmable wallets for AI agents

Cloudflare has launched programmable crypto wallets paired with claimable handles (cloudflare.pay), enabling AI agents to possess verifiable identities on the web. This infrastructure allows autonomous agents to execute micro-transactions seamlessly while offering developers configurable spending controls and policy guardrails to keep agent transactions secure.

MODEL29m ago

OpenAI deploys GPT-Live real-time voice system

OpenAI has deployed GPT-Live, a real-time voice system engineered for fluid, bidirectional verbal communication. Unlike traditional voice models that wait for complete user utterances before generating a response, GPT-Live processes streaming audio to listen, reason, and speak concurrently.

UPDATE31m ago

ElevenLabs has updated Scribe with improved entity detection featuring 17% higher accuracy, nine new entity types, and real-time processing support.

ElevenLabs has upgraded entity detection in Scribe, delivering 17% higher accuracy, nine new entity types, and real-time processing capabilities. The feature detects structured and sensitive data in transcripts down to the exact character position across PII, PHI, and PCI categories covering over 50 entity types. Developers can redact sensitive values using customizable formats, such as traceable numbered tokens, accessible via API parameters or ElevenLabs coding agent skills.