Llama 3.3 price surges 610% as ecosystem adds 86 projects
67,747 stars today86 new projects+3,687 awesome-gpt-image-2
By olud.ai editorial · built from our own tracking data, published daily
Meta's Llama 3.3 70B Instruct input price jumped 610% to $0.71 per 1M tokens, the largest single-day increase among tracked models. DeepSeek V4 Pro rose 55% to $0.87, while Qwen2.5 VL 72B dropped 69% to $0.25. The open-source AI ecosystem gained 67,747 stars across 10,000+ projects today, with 86 new projects entering the index.
This week 1,998 new projects have been added, plus 100 new Hugging Face spaces. The top gainer today is awesome-gpt-image-2 with +3,687 stars. Among emerging projects, archify (17,716 stars) offers diagram generation, and LTX-Desktop (1,922 stars) provides video generation with LTX models.
On the research front, the FrontierChallenge benchmark evaluates scientific workflow completion across multiple domains, highlighting gaps in current agent capabilities. Releases include Windmill v1.797.0 with configurable S3 URL expiry, Langfuse v4.22.0 with evaluator tracking, and Pydantic AI v2.35.1 with Bedrock guardrail fixes.
Releases of the day
Windmill v1.797.0 adds configurable expiry for presigned S3 public URL signatures. Mem0 deepseek-plugin-v0.1.1 introduces anonymous usage telemetry batched to PostHog. Langfuse v4.22.0 tracks evaluator creation configuration and allows serving the web build. Pydantic AI v2.35.1 fixes Bedrock guardrail trace handling in ModelResponse. LoopX v0.5.3 is recommended for all v0.5.2 users, especially those running long autonomous goal continuations. Puppyone v0.3.10 is a signed and notarized desktop release.
Input prices moved in both directions. DeepSeek V4 Pro 0423 rose 55% to $0.87 per 1M tokens, and DeepSeek V4 Flash 0731 rose 50% to $0.06. Llama 3.3 70B Instruct surged 610% to $0.71, and Llama 4 Scout rose 10% to $0.11. On the down side, Gemma 4 31B dropped 10% to $0.09, Qwen3.5 397B A17B fell 22% to $0.39, Qwen2.5 VL 72B Instruct fell 69% to $0.25, and Hy3 fell 38% to $0.08.
▲ 55%DeepSeek V4 Pro 0423DeepSeek · input $0.56 → $0.87 per 1M tokens
Eight new projects were detected today. archify (17,716 stars) generates architecture and workflow diagrams as self-contained HTML. LTX-Desktop (1,922 stars) is an open-source desktop app for LTX video generation. failproofai (1,508 stars) provides observability and policy enforcement for AI agent harnesses. OpenRath (1,097 stars) is a PyTorch-like runtime for dynamic multi-agent workflows. alethe-agents (458 stars) is a local-first desktop workspace for coding agents. anolisa (512 stars) is an agentic OS with runtime and security. humla (249 stars) records and transcribes meeting audio on-device. AndroLLM (132 stars) brings AI to Android with hardware acceleration.
FrontierChallenge evaluates scientific workflow completion across multiple domains. The benchmark tests agents on data analysis, code execution, and research artifact production, revealing that most current benchmarks focus on final answers or isolated programs. The work is by Liangcai Su, Zhaopeng Feng, Zhuo Chen and colleagues, and received 97 upvotes on Hugging Face.
Today's facts show extreme price volatility in popular models, with Llama 3.3 and DeepSeek V4 Pro rising sharply while Qwen2.5 VL and Hy3 fall. The ecosystem continues to expand rapidly, with new projects targeting agent orchestration, observability, and desktop tools. The FrontierChallenge benchmark underscores a growing focus on evaluating agents in realistic, multi-step scientific tasks.
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
