Price swings hit DeepSeek, GLM, OpenAI as 2,177 projects join index
85 new projects
By olud.ai editorial · built from our own tracking data, published daily
DeepSeek V4 Flash input price dropped 29% to $0.05 per 1M tokens, while OpenAI's gpt-oss-120b jumped 275% to $0.15 per 1M tokens. Z.AI cut GLM 5.2 by 61% to $0.55 and GLM 5.3 by 35% to $0.91. Moonshot AI lowered Kimi K3 by 35% to $1.95. Meta raised Muse Glimmer 30B by 17% to $0.35.
The ecosystem added 2,177 new projects this week and 100 new Hugging Face spaces. The index now tracks over 10,000 open-source projects and 300 models with live pricing.
The price moves span a wide range, from aggressive cuts on Chinese models to a steep hike on an OpenAI offering. The research community focused on KV cache compression for DeepSeek's latest flash model.
Releases of the day
OpenClaw v2026.9.5 shipped with 64 direct commits from 503 contributors across 4,179 pull requests. The release notes and changelog contain the same content. AutoGPT released autogpt-platform-beta-v0.8.0 with a breaking change: ENCRYPTION_KEY, UNSUBSCRIBE_SECRET_KEY and BETTER_AUTH_SECRET no longer ship with default values in .env.default. Pydantic AI v2.46.0 added a feature letting TypeSafeModel fill a tool's arguments when Jev can express them.
DeepSeek V4 Flash input fell 29% to $0.05 per 1M tokens. GLM 5.2 dropped 61% to $0.55, and GLM 5.3 dropped 35% to $0.91. Kimi K3 fell 35% to $1.95. On the upside, OpenAI's gpt-oss-120b rose 275% to $0.15, and Meta's Muse Glimmer 30B rose 17% to $0.35. The moves affect users of these models, with the largest absolute cut on GLM 5.2 and the largest percentage hike on gpt-oss-120b.
A Hugging Face community paper titled "DeepSeek-V4.1-Flash: Pushing the Limits of KV Cache Compression" by DeepSeek-AI authors addresses the high cost of prefill in long-context workloads. The work aims to reduce KV cache size, which is a key bottleneck for input-heavy agent tasks. The paper received 80 upvotes on the platform.
Today's facts show a market in flux: aggressive price cuts on several Chinese models contrast with sharp increases on OpenAI and Meta offerings, while the research focus on KV cache compression signals continued pressure to lower inference costs for long-context applications.
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
