Cline

Due to abnormally high abuse, we are pausing the free DeepSeek-V4.1-Flash promotion. We are actively investigating and trying to mitigate the situation.

Mistral Large 4 beats Opus 5.5 and GPT-6 Astra on cybersecurity benchmarks, largely because it refuses far fewer security tasks. Meanwhile Opus and Astra had ~40% of tasks blocked by their own safety filters.

Step 5 Preview is now free in Cline. On DeepSWE it scores ahead of Kimi K3 and GLM-5.3, making it one of the strongest open weights coding models available.

Pareto 26.10 Preview is a new model available in Cline. It routes to several frontier and open models, grades the answer, & returns the best one - while preserving prompt cache + keeping cost low. ~56x cheaper than Fable for the same DeepSWE score at $0.24 per task vs $13.41!

Meta’s Muse Wonder now leads on DeepSWE. This is an incredible turnaround only 14 months after the announcement of Meta Superintelligence Labs. We have a new model on the frontier.

1/ Introducing Cline Cloud Agents. You can now use Cline straight from your browser. Give it a task and it runs in a secure cloud sandbox, writing code, testing its work, then opening a PR on your GitHub repo.

Solar Mini 4 is free in Cline. It’s a new model from Korean lab @upstageai - 524K context, 208 tokens/sec, 35B MoE with only 3B active parameters. On the AAII it scores 24, the highest of any model at 3B active, and within a point of Nemotron 3 Ultra, which uses 55B active!

Gemini 4 Argon is #1 on Val Index, completing agentic tasks at half the cost of Opus 5.5. Google shared that the Chrome team recently used Argon to refactor ~6 million lines of legacy features out of a 40-million line codebase, eliminating an entire category of browser crashes.

DeepSeek V4.1-Flash-1017 has released, beating GPT 5.6 Sol on Terminal-Bench while being ~17x cheaper. It improves the previous DeepSeek V4.1-Flash result by 35%, accomplished with "no algorithmic innovation" but instead by RL training it on multiple coding agent harnesses.

Gemini 4 Carbon is the first model to surpass Claude and GPT, and beats Opus 5.5 on nearly every benchmark at half the cost. That’s an 18% improvement in only 2 weeks since Argon.

Kimi K3.1 Pro nearly matches Opus 5.5 performance on Terminal-Bench 4.0 at *half* the cost. Moonshot also removed cyber data from mid-training and RL to deliberately reduce performance on exploit tasks by ~10%, while everything else, including coding, kept improving.

MiniMax M3.1 Flash goes toe-to-toe with Opus and GPT on coding at ~25x lower cost. This was the model behind Space Bunny, the stealth model that’s been topping usage leaderboards recently. Free in Cline now!