AI News · August 20, 2026

Price moves split as DeepSeek V4 Pro jumps 118% while GLM 5.2 drops 18%

149,514 stars today57 new projects+5,661 public-apis
By olud.ai editorial · built from our own tracking data, published daily

The ecosystem gained 149,514 GitHub stars across 10,000+ tracked open-source AI projects today, with 57 new projects entering the index. This week 2,171 new projects have been added, along with 100 new Hugging Face spaces. The top gainer today is public-apis with +5,661 stars.

Several minor releases shipped across the agent and tool ecosystem. n8n 2.35.5 fixes task runner restart issues for slow runners. CrewAI 1.15.17 adds declarative conversational flows and built-in conversational methods. OpenCode v1.18.19 adds native OpenAI and Anthropic passthroughs for Cloudflare AI Gateway. Stagehand v3.7.5 includes a contribution fix and retargeted release automation. Pydantic AI v2.32.1 rejects synchronous callbacks inside agent runs. llama.cpp b10509 adds ggml_rope_set_offset with metal support.

Releases of the day

n8n [email protected] 2.35.5 (2026-08-20) · Bug Fixes · * core: Avoid restarting task runners that are only slow (#36610) (7ee5295) * core: Only init the expression engine for commands that evaluate... OOpenCode v1.18.19 Core · Improvements · - Added native OpenAI and Anthropic passthroughs for Cloudflare AI Gateway models. - Matched Codex rate limits more closely to ChatGPT subscription... OOpen Interpreter rust-v0.0.40 CrewAI 1.15.17 What's Changed · Features - Add declarative conversational flows documentation - Synthesize built-in conversational methods for declarations - Enable declarations to drive... Stagehand stagehand-server-v3/v3.7.5 What's Changed * Contrib/2384 by @miguelg719 in https://github.com/browserbase/stagehand/pull/2385 * chore(release): retarget release automation from main to v3 by @miguelg719... Pydantic AI v2.32.1 What's Changed 🐛 Bug Fixes * Reject Agent.run_sync() from synchronous callbacks inside agent runs by @dsfaccini in https://github.com/pydantic/pydantic-ai/pull/7277 * Don't... SurfSense v0.0.37 llama.cpp b10509 ggml: add ggml_rope_set_offset (+ metal support) (#27120) · * add params · * cpu kernel · * metal kernel · * add test backend ops · * gate other backends · * ggml: (cuda)...

Pricing moves

Price moves are split today. DeepSeek V4 Pro 0423 input price rises 118% to $1.44 per 1M tokens. Qwen3.5-35B-A3B input price rises 14% to $0.25 per 1M tokens. Qwen3.6 27B input price rises 3% to $0.30 per 1M tokens. On the downside, GLM 5.2 input price drops 18% to $0.97 per 1M tokens. Kimi K2.5 input price drops 21% to $0.45 per 1M tokens. Kimi K2.6 input price drops 4% to $0.54 per 1M tokens. Qwen3 Next 80B A3B Instruct input price drops 10% to $0.09 per 1M tokens. DeepSeek V3 0324 input price drops 7% to $0.25 per 1M tokens.

▼ 18%GLM 5.2Z.AI · input $1.19 → $0.97 per 1M tokens
▲ 118%DeepSeek V4 Pro 0423DeepSeek · input $0.66 → $1.44 per 1M tokens
▼ 4%Kimi K2.6Moonshot AI · input $0.56 → $0.54 per 1M tokens
▼ 21%Kimi K2.5Moonshot AI · input $0.57 → $0.45 per 1M tokens
▼ 10%Qwen3 Next 80B A3B InstructAlibaba · input $0.1 → $0.09 per 1M tokens
▼ 7%DeepSeek V3 0324DeepSeek · input $0.27 → $0.25 per 1M tokens
▲ 14%Qwen3.5-35B-A3BAlibaba · input $0.22 → $0.25 per 1M tokens
▲ 3%Qwen3.6 27BAlibaba · input $0.29 → $0.3 per 1M tokens

Emerging projects

A new project called colleague-skill entered the index today with 23,537 stars. It is described as transforming cold farewells into warm skills, positioning itself as part of a Digital Life 1.0 concept.

colleague-skillPython · 23,537 stars · 将冰冷的离别化为温暖的 Skill，欢迎加入数字生命1.0！Transforming cold farewells into warm skills? It's giving rebirth era....

Research pick

A Hugging Face community paper with 96 upvotes presents Zetta zeta, a closed-loop embodied harness for self-evolving physical intelligence. The authors argue that existing harnesses remain open-loop and follow fixed skills, while their approach enables closed-loop learning in physical execution.

Zetta ζ: An Efficient Closed-Loop Embodied Harness for Self-Evolving Physical IntelligenceXin Ding, Liang Mi, Mingzhe Huang et al. · 96 community upvotes · Embodied agents are increasingly used to close the gap left by end-to-end policy models. Yet the agentic path has not realized closed-loop learning in physical execution: existing harnesses remain largely open-loop, following fixed skills d

The bigger picture

Today's price moves show no clear direction, with increases and decreases spread across providers. The minor releases focus on bug fixes and declarative features, suggesting a maintenance and refinement phase across the agent tooling landscape.

Source: olud.ai tracking of 10,000+ open-source AI projects, 300+ models and live provider pricing. Figures are measured, not estimated. All releases · Live pricing · Latest in AI

← All AI news