DeepSeek marks up two V4 Pro models, Qwen3.6 doubles price
149,514 stars today55 new projects+5,661 public-apis
By olud.ai editorial · built from our own tracking data, published daily
DeepSeek raised prices on two of its V4 Pro models today. The V4 Pro 0423 input price rose 11 percent to $1.60 per million tokens. The newer V4 Pro 0813 input price jumped 80 percent to $1.19 per million tokens. Meanwhile Alibaba's Qwen3.6 27B input price doubled to $0.60 per million tokens. On the other side of the ledger Google cut the Gemma 4 31B input price 10 percent to $0.09 per million tokens.
The ecosystem today added 149,514 GitHub stars across the 10,000-plus tracked open-source AI projects. Fifty-five new projects entered the index. The public-apis repository led all gainers with 5,661 stars. This week 2,176 new projects and 100 new Hugging Face spaces have been added to the index.
The release slate was dominated by minor bug-fix updates. Cline completed a rebrand from Cline Code with refreshed icons. Pydantic AI shipped a breaking change for Anthropic users who had not updated from legacy httpx. Llama.cpp fixed a metal kernel issue for tensor API matrix-matrix multiplication.
Releases of the day
Several projects shipped minor releases today. N8n version 2.35.6 normalises environment variable values before schema-based parsing and resolves end-user credentials when loading node parameters. OpenCode v1.18.20 surfaces failed subagent tool calls with a resumable task ID and retries provider responses that end with finish_reason network_error as well as other network errors. AutoGPT beta v0.7.2 adds an AI-voice narrative to its morning briefing and per-expert spend tracking on home team cards. Cline desktop v0.0.15 renames the app from Cline Code to Cline. Settings, sessions and credentials carry over unchanged. Langfuse v4.16.0 extends signup attribution to LinkedIn, Reddit and X click IDs. Pydantic AI v2.33.0 ships with a warning: Anthropic 1.0.0 rebuilt on httpx2 and reached PyPI on August 20. Users who rely on legacy httpx support will see Anthropic stop working. Llama.cpp b10545 clamps the K extent in the tensor API mat-mat kernel for cases where K is not a multiple of 32.
Four models saw input price changes today. DeepSeek V4 Pro 0423 input price rose 11 percent to $1.60 per million tokens. DeepSeek V4 Pro 0813 input price rose 80 percent to $1.19 per million tokens. Alibaba Qwen3.6 27B input price doubled to $0.60 per million tokens. Google Gemma 4 31B input price fell 10 percent to $0.09 per million tokens. Users of DeepSeek and Qwen models face higher costs, while Gemma users get a discount.
▲ 11%DeepSeek V4 Pro 0423DeepSeek · input $1.44 → $1.6 per 1M tokens
▲ 80%DeepSeek V4 Pro 0813DeepSeek · input $0.66 → $1.19 per 1M tokens
Emerging projects
Seven new projects entered the index today. Omnibase is a self-hosted AI workbench for knowledge, RAG, model providers and governed user-built agents, with 166 stars. Mirage, with 3,545 stars, describes itself as the world's first unified virtual filesystem for AI agents. Aimock, at 888 stars, mocks every external service an AI app talks to — LLM APIs, MCP, A2A, AG-UI, vector DBs and search — in one package with one port and zero dependencies. Anysearch-mcp-server, gaining 1,827 stars, is a unified real-time search MCP server supporting general web search, vertical domain search, parallel batch search and full-page URL fetching. Huashu-design, with 23,342 stars, is an HTML-native design skill for Claude Code that creates high-fidelity prototypes, slides, and animations guided by 20 design principles and a five-dimension review process. Trace-file-lineage, at 521 stars, finds which script, notebook, data, command or AI agent produced a file locally with evidence and honest uncertainty. Agent-plugins-spec v1.0.0, at 1,102 stars, defines a minimal standard for packaging agent extensions into distributable plugins.
A Hugging Face community paper with 175 upvotes introduces EnvHarness, a method that lets LLM agents learn by interacting with environments that adapt to the agent's weaknesses. Current environments are hand-built and static — they do not change as the agent improves. EnvHarness generates new environments on the fly, keeping the learning process challenging rather than letting agents plateau on fixed tasks.
Today's pricing moves suggest a two-tier trend: leading models like DeepSeek V4 Pro and Qwen3.6 are becoming more expensive, while a competitor like Gemma 4 becomes cheaper. The release activity is quiet and maintenance-focused. The seven new projects, especially the high-star huashu-design and mirage, point to growing interest in agent infrastructure — filesystem abstractions, plugin standards, and design tools for code-writing AI.
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
