@karminski3: Finally, a useful local DeepResearch is available! I've packaged a skill for everyone, ready to use out of the box. The driving model is Apodex's newly released open-weight DeepResearch fine-tuned model Apodex-1.0-mini (now available a…

X AI KOLs Timeline Models

Summary

Apodex has released an open-weight DeepResearch fine-tuned model, Apodex-1.0-mini, based on Qwen3.5-35B-A3B. It scores 71.5 on BrowseComp, approaching flagship model performance, and can run efficiently locally. The author packaged an out-of-the-box skill.

Finally, a useful local DeepResearch is available! I've packaged a skill for everyone, ready to use out of the box. The driving model is Apodex's newly released open-weight DeepResearch fine-tuned model Apodex-1.0-mini (now available as Apodex deep research). This model has directly topped the FutureX leaderboard (a benchmark for LLM prediction accuracy created by ByteDance and Stanford)! Now it can be used locally. The model is post-trained on Qwen3.5-35B-A3B. It scores 71.5 on BrowseComp (AIAgent web browsing capability test), approaching the level of flagship models. And it's small enough that my MacStudio can run it at 70+ tokens per second. I was curious about the difference between a DeepResearch fine-tuned model and a regular model, so I integrated Apodex's single-agent DeepResearch framework using this skill and tested it specifically to see if the DeepResearch fine-tuned model is just hype: #Apodex #DeepResearch #ReAct #DiscoverativeIntelligence #FutureX
Original Article
View Cached Full Text

Cached at: 07/07/26, 01:31 PM

Finally, a good local DeepResearch is available! I’ve packaged a skill for you, ready to use out of the box.

The driving model is Apodex’s newly released open-weight DeepResearch fine-tuned model Apodex-1.0-mini (now available as Apodex deep research). This model directly tops the FutureX leaderboard (a benchmark by ByteDance and Stanford for predicting the accuracy of large models’ future predictions)! Now it can run locally.

The model is post-trained on Qwen3.5-35B-A3B, achieving 71.5 points on BrowseComp (AI Agent web browsing capability test), approaching flagship model level. And it’s small enough—my MacStudio runs at 70 tps+.

I was curious about the difference between DeepResearch fine-tuned models and regular models, so I used this skill to integrate Apodex’s single-agent DeepResearch framework and specifically tested it to see if the DeepResearch fine-tuned model is just hype:

#Apodex #DeepResearch #ReAct #DiscoverativeIntelligence #FutureX

Similar Articles

@zhixianio: After receiving the new machine, I began an 'ascetic' practice of forcing myself to use local models for common tasks. I thought it would be painful, but both speed and quality greatly exceeded my expectations: Model: Qwen3.6-35B-A3B-oQ6-fp16-mtp, Running: oMLX, with N…

X AI KOLs Timeline

The author uses the Qwen3.6-35B-A3B model and oMLX tool on the new local machine for daily tasks, finding that both speed and quality far exceed expectations, even outperforming remote LLMs in PA and coding scenarios, demonstrating a significant improvement in on-device AI capabilities.

@RookieRicardoR: Domestic models break through again, matching top models like Claude 4.6 and Gemini 3.1 Pro. Just tested Qwen3.7-Max, sharing some real thoughts. Last night I topped up as soon as the API went live and chose three tasks (see video) to test Qwen3.7-Max's frontend capabilities…

X AI KOLs Timeline

The user tested Qwen3.7-Max and believes it matches top models like Claude 4.6 and Gemini 3.1 Pro in frontend, computing power, and Agent capabilities. Its reasoning ability has significantly improved, and with monthly iteration speed, it has become a first-tier domestic model.

@yidabuilds: https://x.com/yidabuilds/status/2053409619641602286

X AI KOLs Timeline

The author conducted a comparative evaluation of four domestic AI models: DeepSeek V4, Kimi K2.6, GLM-5.1, and MiniMax M2.7. The analysis covers their strengths and weaknesses regarding cost, long-context processing, coding stability, and reasoning performance, offering specific recommendations on how to route tasks involving large document analysis, long-running background jobs, and bulk content generation.

@yibie: Recommend this project. shadcn (creator of shadcn/ui) built an Agent Skill — let the most expensive model handle auditing and planning, and let the cheap model write code. The idea itself isn't new, but shadcn turned it into an installable, orchestrated system with an execution closed loop. Every design decision is practical...

X AI KOLs Timeline

shadcn released the Agent Skill project improve, which has high-cost models perform code auditing and planning while low-cost models execute, forming an installable, orchestrated system with an execution closed loop.