DeepSeek and OpenAI slash prices as LiteLLM signs images
81 new projects
By olud.ai editorial · built from our own tracking data, published daily
DeepSeek cut the input price of its V4 Flash 0423 model by 17% to $0.05 per 1M tokens, while OpenAI reduced gpt-oss-20b by 33% to $0.02 per 1M tokens. The moves add to a week where 2,309 new projects entered the index and 100 new Hugging Face spaces appeared. The pricing landscape continues to shift rapidly, with seven models seeing price changes today alone.
Alibaba raised Qwen3.6 27B input price by 7% to $0.32 per 1M tokens, bucking the downward trend. Z.AI adjusted multiple GLM 5.3 variants: the standard model fell 29% to $0.65, the Flash variant rose 87% to $0.15, and the Flash batch version dropped 25% to $0.06. Meta trimmed Llama 4 Maverick by 5% to $0.19 per 1M tokens.
Releases of the day
LiteLLM v1.102.1 ships a security improvement: all Docker images are now signed with cosign using a key introduced in commit 0112e53. Users can verify image integrity before deployment. Langfuse v4.42.0 refactors web imports and fixes the trace page, improving internal code organisation and UI stability. Pydantic AI v2.48.0 adds support for genai-prices 0.1.7 and new OpenAI audio models, plus the OpenAI gpt-6-sol model. Puppyone v0.3.25 delivers a signed and notarised desktop release, promoted from internal version v0.3.25-internal.46.
Seven models changed input prices today. DeepSeek V4 Flash 0423 dropped 17% to $0.05, DeepSeek V4 Pro 0423 fell 3% to $0.87, and OpenAI gpt-oss-20b fell 33% to $0.02. Alibaba raised Qwen3.6 27B by 7% to $0.32. Meta cut Llama 4 Maverick by 5% to $0.19. Z.AI lowered GLM 5.3 by 29% to $0.65 and GLM 5.3 Flash batch by 25% to $0.06, but raised GLM 5.3 Flash by 87% to $0.15.
A Hugging Face community paper titled "RULER: Instance-aware Rubric Rewards for SVG Generation" proposes a new method for evaluating and optimising SVG code generation from natural language. The authors argue that existing scalar metrics like CLIP score are unreliable for this open-ended task, and introduce rubric-based rewards tailored to each generated instance.
The combination of widespread price cuts and new security-focused releases suggests the ecosystem is maturing toward lower cost and higher trust, even as individual model pricing remains volatile.
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
