While DeepSeek V4-Flash is significantly cheaper on price per token, this can be misleading if the overall cost per task ends up being higher due to more turns being made. However, @ArtificialAnlys reports DeepSeek completing the same benchmark tasks as Fable at 105x lower cost.

DeepSeek silently updated their changelog with a new V4-Flash upgrade 1 hour ago. Their new Terminal-Bench score is 82.7, a massive +25.8 point leap from its initial April preview score of 56.9. Currently only available via their API, open weights release will follow shortly.

We had Kimi K3 recursively self-improve the Cline harness to improve its own performance. 17 hours later, it went from 77.5% to 88.8% on Terminal Bench, and cut run cost from $79 to $49.8.

Kimi costs ~3-12x cheaper than Fable, but how much more could you save hosting it yourself? We ran the numbers on Cline’s production traffic, and the results: ~10% savings, 25%+ with time-of-day autoscaling (but this only works if over $500K/year of spend) We predict self-hosting open weight models is going to become standard practice as businesses scale their token consumption with increased adoption, given the price and data sovereignty benefits. And we’re excited to see Kimi K3 bring us closer to this future, enabling broader and more competitive access.

We’ve been impressed with GLM-5.2 and so are introducing a $9.99/month subscription to give you 2-5x discounted access to it and other open weight models like DeepSeek, Kimi, MiniMax, Mimo, Qwen. Use it on Cline CLI & IDE with $1.99 special promo if sign up via: npm i -g cline

We’re getting together with @Zai_org for a FIFA World Cup watch party ⚽️ USA vs. the world on a big screen, food and drinks on us, and a fireside chat on open-weight models and where agentic coding is headed. Come join us!

Run Cline on Local AI models with Atomic Chat! @cline is a coding agent trusted by 8M+ developers. Write, refactor, ship code securely on your own hardware with local models powered by @atomic_chat_hq — no cloud, private, free and open-source

We've kept hearing how GLM-5.2 beats Opus 4.8, and are skeptical of benchmarks - so we tested them on a real bug from the Cline repo. While both models fixed the issue, GLM was the winner in terms of cost and code quality: - GLM used twice as many tokens (GLM 1.1m vs Opus 660K) but cost half as much (GLM $0.41 vs Opus $0.81) - Opus finished quicker - 1.6 min and 12 tool calls vs GLM 4.7 min and 28 tool calls - GLM cleaned up dead code and verified the build compiled before completing. Opus didn't - it left type errors that passed tests but broke the production build. Both runs used the same Cline harness prompting and tools, so it seems GLM is RL trained to spend more tokens verifying its work before completing. Impressive work by the @Zai_org team!

so many golden nuggets from the glm 5.2 release blog about breakthroughs that helped them with benchmark gains. you never see this level of transparency from the frontier labs. they found that glm 5.2 kept trying to reward hack in rl by curl'ing task related source from github repos, and grep'ing for eg "*hidden*" or "secret_cases.json" fishing around its sandbox for files it wasnt supposed to have access to and try to find answers. they mitigated this by using an llm judge to check the intent of tool calls that matched a list of suspicious tool call patterns. if a hack was detected, the system blocked the grep/curl/etc and returned dummy information as a result. importantly this allowed the model to continue working instead of rejecting and interrupting the entire trajectory, which helped prevent training instability.

Introducing the Cline SDK. We rebuilt the Cline harness for our extension and CLI from scratch using all the lessons learned since creating one of the world's first coding agents in 2024, and are open sourcing it for others to build with today. npm i @​cline/sdk 🧵

We also added free access to Kimi K2.5, Moonshot's open-source model. Strong on agentic tasks, way cheaper than closed models. Free access is temporary, so make the most of it!