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

Artificial Analysis@ArtificialAnlys·Collected 62d ago

👀87

🔥94

Artificial Analysis: Muse Spark 1.3 hits Index 62 with leading agentic cost efficiency

Artificial Analysis (@ArtificialAnlys) benchmarks Meta Muse Spark 1.3: the xhigh variant scores 61 on the Intelligence Index (up 4 from 1.2’s 57), while limited-preview max reaches 62, trailing only Claude Fable/Opus lines. Gains are mainly agentic and scientific; xhigh is the lowest cost-per-task among models at 59+ (~$0.55) at unchanged $1.25/$4.25 per 1M tokens.

Key Takeaways

xhigh scores 61; limited-preview max hits 62—up from Muse Spark 1.2’s 57.
Agentic gains on Tau3-Bench Banking, Terminal-Bench, and GDPval; max leads Banking-class results.
Cost: xhigh ~$0.55/task, lowest among 59+ Index models; 1M context, multimodal, via Meta API / Muse Code.

Read full story

Linked Models & Agents:

⚡ Muse Spark 1.3 🤖 Cursor Composer (Agent Mode)🤖 Claude Code (Anthropic)🌐 Meta AI

#Muse Spark 1.3#Artificial Analysis#Benchmark#Agentic#Meta#Intelligence Index

Composio@composio·Collected 65d ago

👀62

🔥89

View on 𝕏 ↗

Composio Evaluates 5 Open-Weight Models on 30 Multi-Step Agent Tasks: Flash Models Lead in ROI

Composio benchmarked 5 open-weight models across 30 multi-step agent tasks, finding GLM-5.3 Flash and DeepSeek-V4 Flash delivered the highest cost efficiency at a fraction of full-model pricing.

Key Takeaways

Flash-class models achieve near parity with full-scale weights at approximately 75% lower inference cost;
Multi-application tool orchestration remains the primary failure mode across all open-source models;
Provides empirical cost-per-success metrics to guide architectural selection for enterprise agent deployments.

Read full story

Linked Models & Agents:

⚡ GLM-5.3-Flash ⚡ DeepSeek-V4-Flash

DeepSeek@deepseek_ai·Collected 65d ago

👀47

🔥99

View on 𝕏 ↗

DeepSeek-V4 Preview is live and open-sourced with 1M context and frontier agentic coding SOTA

DeepSeek launched and open-sourced V4 Preview: V4-Pro (1.6T total / 49B active) rivaling closed-source frontier models, and V4-Flash (284B / 13B active) for cost efficiency. The model achieves open-source SOTA on agentic coding with default 1M context and day-one integration across Claude Code and OpenCode.

Key Takeaways

V4-Pro features a 1.6T/49B MoE architecture with open-source SOTA on agentic software development.
1M context window is now the standard default across official DeepSeek web and API endpoints.
API is fully OpenAI/Anthropic compatible, providing drop-in integration for coding agents.

Read full story

Linked Models & Agents:

⚡ DeepSeek-V4 (Next-Gen MoE)⚡ DeepSeek-R1 (Open Reasoning)🤖 Claude Code (Anthropic)🌐 DeepSeek

#Deepseek#Deepseek v4#Open source#Moe#1m context#Agentic coding