AI News · September 1, 2026

DeepSeek V4 Pro price surges 145%; new projects hit 4K stars

57,684 stars today67 new projects+3,433 archify

DeepSeek V4 Pro input price rose 145% to $1.03 per 1M tokens, the largest price move today. The increase follows a period of competitive pricing among frontier models and signals a shift in DeepSeek's cost strategy. The overall ecosystem gained 57,684 GitHub stars across 10,000-plus tracked projects today, with 67 new projects entering the index and 1,889 new projects added this week. Weekly additions also include 100 new Hugging Face spaces.

Two other models saw price adjustments. Kimi K2.5 input price dropped 25% to $0.45 per 1M tokens, while Qwen3 Next 80B A3B Instruct increased 11% to $0.10 per 1M tokens. These moves keep the model pricing landscape in flux, with some providers lowering costs and others raising them. The top gainer on GitHub today is archify with +3,433 stars.

Multiple emerging projects crossed 1,000 stars, including sprix-sage-router at 2,961 stars and x64dbg-mcp-server at 1,800 stars. The largest new project detected today is maka at 4,316 stars, an Apache incubating local-first AI agent workspace. These projects span agent routing, debugger integration, 4D video generation, and AI education.

Releases of the day

Seven open-source projects shipped minor releases today. Langflow v1.11.6 fixed MCP public variables and stdio headers. LiteLLM v1.99.0 introduced cosign-based Docker image signature verification. Playwright MCP v0.0.80 added browser recording tools via an opt-in devtools capability. Pydantic AI v2.37.0 added support for glm-5.3-flash and reworked its Z.AI test suite. Phoenix MCP v4.3.5 updated dependencies without adding new tools. LMDeploy v0.17.0 integrated DeepEPv2 and added support for Kimi K2.6. Vespa v8.748.3 updated its command-line tool for the Vespa.ai platform. For users, these releases improve security verification, extend model compatibility, and add debugging and recording capabilities to agent workflows.

Pricing moves

DeepSeek V4 Pro input price increased 145% to $1.03 per 1M tokens. Kimi K2.5 input price fell 25% to $0.45 per 1M tokens. Qwen3 Next 80B A3B Instruct input price rose 11% to $0.10 per 1M tokens. These changes affect developers using these models for inference, with DeepSeek becoming significantly more expensive and Kimi becoming more affordable.

▲ 145%DeepSeek V4 Pro 0423DeepSeek · input $0.42 → $1.03 per 1M tokens
▼ 25%Kimi K2.5Moonshot AI · input $0.6 → $0.45 per 1M tokens
▲ 11%Qwen3 Next 80B A3B InstructAlibaba · input $0.09 → $0.1 per 1M tokens

New models

Two new model variants were added to existing frameworks. LMDeploy now supports Kimi K2.6 with its pytorch backend. Pydantic AI added glm-5.3-flash to its provider integrations. These additions expand the model options available within those ecosystems.

Emerging projects

Eight projects were detected today. Sprix AI router (sprix-sage-router) with 2,961 stars offers state-aware routing for A2A agent networks. Maka with 4,316 stars is an Apache incubating local-first AI agent workspace. 4DAnyone with 1,046 stars creates 4D avatars from monocular video. x64dbg-mcp-server with 1,800 stars exposes debugger functionality over MCP. Other projects include a 24-week AI engineering course (245 stars), a Databricks rendering tool (129 stars), a DeepSeek harness plugin (105 stars), and a photo-to-companion tool (200 stars).

Research pick

The paper GenFirst proposes generating latent representations before reconstruction in latent generative models. It argues that traditional two-stage pipelines optimise for reconstruction rather than generation, and that generating first improves stability in end-to-end latent modeling. The work received 50 upvotes on Hugging Face.

The bigger picture

Today's price moves show a market diverging: DeepSeek is raising prices while Moonshot cuts them, suggesting different competitive strategies. New projects around agent routing and local-first workspaces indicate growing demand for infrastructure that manages multi-agent interactions. The steady inflow of projects and stars suggests the AI open-source ecosystem continues to expand rapidly.

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