DeepSeek API pricing, updates and model releases
Briefpanel reads DeepSeek's API pricing page, its API change log, the deepseek-ai organization on Hugging Face and the deepseek-ai organization on GitHub once a day and records what changed: a new model, a price that moved, a new release. Every item below comes straight from DeepSeek's own pages and shows when that page was last checked.
4 sources17 changes in the last 7 days
Get an alert when DeepSeek changes a price
Get an alert when something changes on the page you pick.
No credit card required · Set up in 10 minutes
Fri, Oct 30
17 itemsNew
DeepEP
DeepEP (DeepEveryParallel) is a high-performance communication library for machine learning training and inference, providing high-throughput and low-latency expert-parallel all-to-all GPU kernels for MoE dispatch and combine, including FP8 dispatch, plus NVLink weight and gradi…
New
DeepGEMM
DeepGEMM is a unified, high-performance tensor core kernel library that brings together key computation primitives of modern large language models into a single cohesive CUDA codebase. It includes GEMMs (FP8, FP4, BF16), fused MoE with overlapped communication (Mega MoE), MQA sc…
New
DeepInfer
DeepInfer is DeepSeek's high-performance, large-scale inference framework, released as open source to provide offline inference with the core features needed for serving at scale. It supports the DeepSeek-V4.1-Flash model and runs on NVIDIA SM100 GPUs and Huawei Ascend 950 NPUs.…
New
deepseek-harness
DeepSeek Harness (dsh) is an open-source agent harness developed by DeepSeek AI, built on an everything-is-a-plugin architecture and powered by Cordis, whose design is described in A Programming Paradigm for Spatiotemporal Composability. It is in developer preview and iterating…
New
FlashMLA
FlashMLA provides efficient multi-head latent attention kernels for large language model inference. Hosted by deepseek-ai, the repository publishes release notes documenting kernels for sparse attention, attention kernels for DeepSeek model versions, and performance updates for…
New
DeepSeek-V4.1-Flash: Pushing the Limits of KV Cache Compression
DeepSeek-V4.1-Flash is a multimodal Mixture-of-Experts model with a Causal Encoder-Decoder architecture that targets the compute, storage, and bandwidth costs of long-context agent workloads. It carries a 552B-parameter backbone and handles contexts up to one million tokens, act…
New
DeepJIT
A lightweight library for xPU kernel JIT compilation
New
TileKernels
A kernel library written in tilelang
New
DeepGEMM-Ascend
DeepGEMM-Ascend: clean and efficient matrix multiplication kernel library for Huawei Ascend NPUs
New
Date: 2024-07-24
deepseek-coder upgraded to DeepSeek-Coder-V2-0724.
New
Date: 2025-09-29
DeepSeek-V3.2-Exp upgrade for deepseek-chat and deepseek-reasoner.
New
Date: 2024-06-14
deepseek-coder upgraded to DeepSeek-Coder-V2-0614 with significantly enhanced coding capabilities.
New
Date: 2025-09-22
DeepSeek-V3.1-Terminus upgrade with language consistency and agent capability improvements.
New
Date: 2025-05-28
deepseek-reasoner upgraded to DeepSeek-R1-0528 with enhanced reasoning, optimized front-end development, reduced hallucinations, and JSON output support.
New
Date: 2025-01-20
deepseek-reasoner is the new model DeepSeek-R1.
New
Date: 2026-08-13
DeepSeek-V4-Pro update: GA release rolled out on APP, Web, and API with significantly enhanced agent capabilities, native support for Responses API, more flexible thinking effort control, and API pricing adjustment.
New
Date: 2026-07-31
DeepSeek-V4-Flash update: official release of the API in public beta with significantly enhanced agent capabilities and native support for the Responses API.
Sat, Oct 17
2 itemsUpdated$0.003
deepseek-flash
What changedMODEL VERSIONDeepSeek-V4.1-FlashDeepSeek-V4.1-Flash-1017DeepSeek-V4.1-Flash-1017 model with 1M context length and 384K max output. Supports non-thinking and thinking modes, JSON output, tool calls, Responses API, Anthropic API, chat prefix completion, and FIM completion (non-thinking mode only); vision supported. Off-peak: cache hit…
Updated$0.022
deepseek-v4-pro
What changedTitleDeepSeek-V4-Pro-0813deepseek-v4-proDeepSeek-V4-Pro-0813 model with 1M context length and 384K max output. Supports non-thinking and thinking modes, JSON output, tool calls, Responses API, Anthropic API, chat prefix completion; FIM completion in non-thinking mode only; vision not supported. Off-peak: cache hit $0.…
How this radar works
Straight from the source
Every item is read from the source page and links back to it. Summaries are automatic; when in doubt, the source is what counts.
Automatic checks
Every page is checked automatically and often. The time of the last check shows on each source.
What changed, in one line
When an item that was already published changes, it shows as Updated, with the difference summarized.
The source page prevails. This summary may be out of date or incomplete: always check the source before acting.
Track the pages that matter to you
BriefPanel reads the page you pick, as often as you choose, and shows the before and after of every field that changed.
No credit card required · Set up in 10 minutes