Ling 3.1 Flash is now in Cline and free until October 13. This 560B total parameter MoE model uses 25B active params, and is on par with other frontier open weights models like Kimi K3 and DeepSeek V4 Pro.

1/ Cline Desktop now has Connectors (beta). Connect Gmail, Slack, Google Calendar, Linear, Sentry, Notion and more in one click. Cline gets tools to use these apps to retrieve context and act on your behalf.

2/ Example: catch up on overnight emails and Slack threads. Cline uses the Gmail and Slack connectors to pull emails and notifications while you make some coffee. “Give me a brief on everything I missed last night, and send replies for anything urgent after I approve”

Two of the largest tasks in Cline the last 30 days: - 9B tokens on Claude Opus 5: ~$8,500 - 9B tokens on DeepSeek V4 Pro: ~$300 That's about 30x cheaper for the same token count. You don't need to spend thousands or wait on limit resets to get large-horizon work done anymore.

Cline Desktop is now available on Linux. (This is an early beta so feedback is welcome! 🙏) Use with ClinePass and all our free models like DeepSeek-V4.1-Flash, Space Bunny Alpha, or BYOK with any provider!

Cline Desktop is open source and built for open-weight models, including local. Run Ollama or LM Studio on the same box, point Cline at it, and the whole loop stays on your machine. Or access the best open weights model at ~5x discount with ClinePass: cline.bot/cline-pass

GPT-6.1 Sol is now in Cline. On DeepSWE v1.1, GPT‑6.1 Sol matches GPT‑6 Astra at roughly 1/5th of the cost!

Sonnet 5.5 beats Opus 5.5 on Terminal-Bench 4.0 at half the cost. It uses far fewer tokens than previous models to do the same work, showing costs reduced up to 30% per task. It also generates output 30% faster than its predecessor, making it the fastest Sonnet model to date.

Ember-1 by the Fireworks Research team is built on Kimi K3, and uses ~40% fewer tokens while achieving the same performance on benchmarks. This was accomplished by post-training K3 to think less repetitively. Reasoning models spend most of their output tokens (sometimes 90%+) on thinking before they answer, which gets expensive in agentic loops where the model tends to re-think the same thoughts on every step. Fireworks RL trained on real agentic coding task loops to teach the model which reasoning actually changes the answer vs which is just looping. In a live A/B test on coding traffic, Ember-1 used 71% fewer reasoning tokens and 39% fewer total tokens than K3 at the same success rate.

This is the first model out of the new Fireworks Research team - it’s wonderful to see an inference provider doing post-training on open weights to reduce costs. Great work @FireworksAI_HQ 💪 fireworks.ai/blog/ember-1