DeepSeek slashes price 50% as OpenAI and Google hike rates
158,487 stars today51 new projects+28,712 rea
By olud.ai editorial · built from our own tracking data, published daily
DeepSeek cut the input price of its V4 Flash 0731 model by 50% to $0.01 per 1M tokens, the largest single drop among eight price moves today. OpenAI raised gpt-oss-120b by 275% to $0.15, Google hiked Gemma 4 31B by 122% to $0.20, and Z.AI increased GLM 5 by 67% to $1.00. Moonshot AI lowered Kimi K3 by 42% to $0.29, Alibaba dropped Qwen3.6 27B by 22% to $0.25, and Z.AI cut GLM 5.1 by 27% to $0.97. Mistral AI raised Mistral Small 3.2 24B by 12% to $0.09.
The tracked open-source AI ecosystem gained 158,487 GitHub stars today, with 51 new projects entering the index. The top gainer "rea" added 28,712 stars. This week 2,516 new projects joined the index, and 100 new Hugging Face spaces appeared.
Releases of the day
Docling v2.140.0 added prefetching of native Distil-Whisper presets and parsing of bold, italic and monospace in AsciiDoc paragraphs. Langfuse v4.61.0 unified manual trace processing with topic generation. OpenCode v1.19.2 locked the 1.x release line so update checks never offer a 2.x build. n8n 2.43.7 suspends in-flight production executions at worker shutdown. Cline desktop-v0.0.47 fixed a plugin loading failure caused by missing SDK packages. Pydantic AI v2.57.0 attached namespaced origin and tool_call_id metadata to BackgroundTools completions. Phoenix v20.23.0 exports server logs over OTLP and forwards Bedrock tool results directly as text.
Eight models saw input price changes. DeepSeek V4 Flash 0731 fell 50% to $0.01 per 1M tokens. Kimi K3 dropped 42% to $0.29. GLM 5.1 decreased 27% to $0.97. Qwen3.6 27B declined 22% to $0.25. Mistral Small 3.2 24B rose 12% to $0.09. Gemma 4 31B increased 122% to $0.20. GLM 5 went up 67% to $1.00. gpt-oss-120b surged 275% to $0.15.
▲ 275%gpt-oss-120bOpenAI · input $0.04 → $0.15 per 1M tokens
Eight new projects were detected. geo-sleuth (2,754 stars) identifies photo locations using OpenStreetMap, elevation skylines, satellite imagery and street view. AgentGarten (183 stars) creates code worlds for evolving agents. lcu (1,096 stars) decouples Codex computer use for any harness. DSHA (1,218 stars) runs DeepSeek Harness on phones without root or Termux. box (402 stars) sandboxes AI agents by restricting execution, read, write and network access. kotlin-footguns (1,287 stars) maps Kotlin and Compose Multiplatform pitfalls from production code. PilotDeck (4,346 stars) is a task-oriented AI agent productivity platform. munder-difflin (8,759 stars) offers an open-source office of Claude Code/Codex-like agents running locally.
OmniCompass, a benchmark for omni-modal video grounding and understanding, was published on Hugging Face with 226 upvotes. It evaluates multimodal large language models across multiple video understanding dimensions that existing benchmarks only partially cover.
Today's price moves show aggressive competition at the low end, with DeepSeek and Moonshot AI cutting rates while OpenAI and Google push higher. The ecosystem continues to expand rapidly, and new projects emphasise agent safety, geolocation, and local deployment, pointing toward a maturing toolchain for production AI agents.
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
