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

👀76

🔥92

Cline CLI 3.0.67 / SDK 0.0.89 / Desktop 0.0.40: pageable MCP oversized results; custom providers actually run tasks

Cline shipped cli-v3.0.67, sdk/sdk/v0.0.89, and desktop-v0.0.40 on 2026-09-30: oversized MCP/Composio tool results are cached with a bounded preview plus a cline://cache/... URI pageable via read_files (16 MiB/session, expire after five unread model iterations); custom providers from providers.json/models.json now resolve on the agent path instead of failing with Unknown or disabled provider; recommended catalog adds GPT-6.1 Sol and refreshes several gateway defaults.

Key Takeaways

Trio: CLI 3.0.67 (23:42:02Z), SDK 0.0.89 (23:34:22Z), Desktop 0.0.40 (23:56:13Z) on 2026-09-30
Oversized MCP/Composio: preview + cline://cache/... pageable via read_files; opt-in resultPolicy: "cache-oversized"; 16 MiB/session, expire after 5 unread iterations
Custom providers now register via resolveGatewayProviderRegistration(Sync) on the agent path
Catalog: GPT-6.1 Sol recommended; refreshed defaults for 302.AI/NanoGPT/Vivgrid/Vultr; Vultr id renames may need re-pin
Desktop remembers Cloud vs Local + Cloud model; MCP settings honor CLINE_* path env vars consistently

Read full story

Linked Models & Agents:

⚡ GPT-6.1 Sol ⚡ Claude Sonnet 5.5 🤖 Cline (VS Code Agent)🤖 cline-sdk 🤖 cline-cli 🌐 cline 🌐 OpenAI 🌐 Anthropic

#Cline#CLI#SDK#Desktop#MCP#Composio#GPT-6.1 Sol#Custom providers#Tooling

Ahmad Awais@MrAhmadAwais·Collected 48d ago

👀61

🔥89

View on 𝕏 ↗

Command Code leads Composio agent-harness results with 18/30 tasks and 375-second average time

Command Code CEO Ahmad Awais shared Composio benchmark results claiming Command Code solved 18 of 30 tasks with a 375-second average time, leading the comparison on open models. The post references Composio testing GPT-6 Astra across six coding-agent harnesses and highlights substantial token differences when harnesses fail.

Key Takeaways

Command Code solved 18 of 30 tasks, with the post claiming the best accuracy among the tested harnesses.
Its reported average task time was 375 seconds, also described as fastest and ahead of DeepSeek's own harness.
The cited Composio test covered Codex, Claude Code, OpenCode, Hermes Agent, Pi Agent, and Command Code.

Read full story

Linked Models & Agents:

#Command Code#Composio#agent harness#Benchmark#DeepSeek#GPT-6 Astra

ComposioDevs@ComposioDevs·Collected 59d ago

👀36

🔥86

View on 𝕏 ↗

Composio: Claude Code plugin composio-plugin-cc v0.2.4 self-updates from SessionStart hook

Composio released composio-plugin-cc v0.2.4 for Claude Code: the plugin can check and update itself from its SessionStart hook, cutting manual upgrade friction so Composio tooling stays current inside sessions.

Key Takeaways

Version: composio-plugin-cc v0.2.4 for Claude Code.
Capability: self-update from SessionStart hook.
Source: official Composio GitHub release.

Read full story

Linked Models & Agents:

🤖 Claude Code (Anthropic)

#Composio#Claude Code#Plugin#SessionStart#Self update

Composio@composio·Collected 65d ago

👀34

🔥87

View on 𝕏 ↗

Composio Expands Unified Tool Layer to 1,384 Apps with 100K Free Monthly Tool Calls

Composio positioned itself as the universal application router for MCP, supporting 1,384 integrated services and providing a 100K free tool calls per month tier for autonomous agent builders.

Key Takeaways

Standardizes multi-app agent tooling into a unified routing and OAuth management layer;
100K free monthly tool calls significantly lowers experimentation barriers for agent developers;
Monitors live production tool-call velocity across more than 1,300 supported SaaS platforms.

Read full story

Linked Models & Agents:

Composio@composio·Collected 65d ago

👀64

🔥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