AI Coding Live News — Real-Time 𝕏/Twitter Updates & Breakthroughs

👀56

🔥90

Cognition: Devin ships Fable 5.1; Fusion matches it on FrontierCode at ~47% lower cost

Cognition announced Claude Fable 5.1 in Devin. With Anthropic’s cache pricing change, Fable-level intelligence is about 54% cheaper—even undercutting Opus. Devin’s Fusion harness matches Fable 5.1 on FrontierCode at roughly 47% lower cost, folding the new model into production agent workflows.

Key Takeaways

Devin now runs Claude Fable 5.1 for coding agents.
Cache pricing makes Fable-class capability ~54% cheaper, undercutting Opus on cost.
Fusion matches Fable 5.1 on FrontierCode at ~47% lower cost for scaled agent runs.

Read full story

Linked Models & Agents:

🤖 Devin 2.0 (Cognition)🌐 Anthropic

#Devin#Fable 5.1#Fusion#FrontierCode#Cognition#Claude#Agent#Pricing

DeepInfra@DeepInfra·Collected 67d ago

👀62

🔥91

View on 𝕏 ↗

DeepInfra Serves GLM-5.3 on NVIDIA Blackwell with Zero Data Retention and Explicit Cache Pricing

DeepInfra deployed GLM-5.3 on day zero utilizing NVIDIA Blackwell infrastructure in the US with Zero Data Retention (ZDR), publishing transparent pricing with an ultra-low $0.26/1M cached token tier.

Key Takeaways

Leverages US NVIDIA Blackwell hardware to maximize inference throughput and minimize time-to-first-token;
Explicit caching tier at $0.26/1M tokens drastically lowers operational costs for iterative agent loops;
Zero Data Retention (ZDR) guarantee provides stringent privacy compliance for enterprise codebases.

Read full story

Linked Models & Agents:

⚡ GLM-5.3