DeepSeek V4-Pro
PulseAugur coverage of DeepSeek V4-Pro — every cluster mentioning DeepSeek V4-Pro across labs, papers, and developer communities, ranked by signal.
- instance of DeepSeek-V4 Flash 90%
- used by DeepSeek-V4 Flash 90%
- used by Claude Opus 4-8 90%
- developed by DeepSeek Chat 90%
- developed by Fireworks AI 90%
- affiliated with deepseek-reasoner 90%
- used by Ascend 910C 90%
- uses DeepClaude 90%
- developed GB300 NVL72 90%
- used by Huawei Ascend 90%
- uses Nexotao 90%
- used by Tencent 90%
- 2026-10-13 product_launch The price of DeepSeek V4 Pro has increased by 95%. source
- 2026-09-12 product_launch DeepSeek is retiring its DeepSeek V4-Pro model and rerouting all requests to the V4.1-Flash variant. source
- 2026-09-09 product_launch DeepSeek has reportedly discontinued its DeepSeek V4 Pro model. source
- 2026-09-05 product_launch DeepSeek has launched its V4 Pro model. source
- 2026-08-31 product_launch DeepSeek V4 Pro experienced significant price volatility, with reports of both a price cut and a subsequent surge. source
- 2026-08-31 product_launch DeepSeek V4 Pro announced a 22% price cut. source
- 2026-08-18 product_launch DeepSeek V4 Pro API prices increased significantly, particularly for cache hit inputs. source
- 2026-08-16 product_launch DeepSeek launched its V4-Pro model with enhanced agent capabilities and a significant price increase. source
- 2026-08-13 product_launch DeepSeek has launched its V4-Pro model and open-sourced its agent software Harness v0.1. source
- 2026-08-13 product_launch DeepSeek V4-Pro and Grok 4.6 models were identified, marking new releases in the LLM space. source
- 2026-08-13 product_launch DeepSeek is launching its V4 lineup of AI models and updating API pricing. source
- 2026-08-13 regulatory DeepSeek V4 Pro's input price increased by 86%. source
- 2026-08-13 product_launch DeepSeek has officially released the DeepSeek-V4-Pro model. source
- 2026-08-13 product_launch DeepSeek officially released its V4 Pro model on August 12, 2026, featuring significant improvements in agentic coding benchmarks. source
- 2026-08-12 product_launch DeepSeek launched its V4-Pro-0813 model with a one-million-token context window and agent-developer tooling. source
26 day(s) with sentiment data
-
AI agent costs heavily influenced by prompt cache hit rates
A recent analysis explores the significant cost implications of prompt caching for AI agents, demonstrating how cache hit rates directly impact billing. The study found that a 20% increase in cache hit rate, from 70% to…
-
DeepSeek V4-Pro and Kimi-k3.1-pro show delivery differences on complex prompts
A comparison of the DeepSeek V4-Pro and Kimi-k3.1-pro models revealed differences in their response delivery, particularly in handling problematic prompts. While both models successfully identified contradictory ticket …
-
Open MoE Models Diverge on Routing Details: DeepSeek, Qwen, Llama 4 Compared
A comparison of open Mixture-of-Experts (MoE) models reveals significant divergence in their routing mechanisms, despite a shared core idea. Models vary widely in expert count, ranging from 8 to 384, and the number of e…
-
MoE Routing Strategies Explained: Top-K, Expert-Choice, Sigmoid Gating and Dynamic Routing
Mixture-of-Experts (MoE) routing strategies are crucial for the performance, load balancing, and cost of large language models. This guide details seven production strategies, including Top-K, Expert-Choice, and Sigmoid…
-
New Unmuzzled AI chat app runs offline, locally on user hardware
A new open-source AI chat application called Unmuzzled has been developed, designed to run entirely offline on a user's local hardware. The application aims to provide a private and user-friendly experience, requiring o…
-
Mo Xin AI secures nearly 1B RMB for sparse inference chips
Mo Xin AI has successfully raised nearly 1 billion RMB in its C+ funding round, marking its second major funding in five months. The company specializes in dedicated sparse inference chips, a technology gaining traction…
-
Open LLMs vs. Claude: Cost vs. Capability Showdown
A comparison of open-source LLMs against Anthropic's Claude models revealed that while open models like GLM 5.3 can be cost-effective for tasks such as data extraction and simple coding, they fall short in complex reaso…
-
FreePi AI agent uses Fusion to route tasks between DeepSeek V4 models
The FreePi AI coding agent has introduced a "Fusion" feature that intelligently routes tasks between two DeepSeek models. DeepSeek V4 Flash is used for initial code exploration and read-only operations, while the more p…
-
LLM API prices mostly fall, but Google raises rates frequently · 1 source tracked
A recent analysis of LLM API pricing changes since 2024 reveals that while prices generally decrease, some models have seen significant increases, particularly in less-scrutinized pricing tiers. Google has been the most…
-
Cheaper AI models like Kimi show cost advantage over GPT-5.5 in bug detection
A recent analysis by Ramp Labs compared the effectiveness of different AI models in identifying security vulnerabilities. When given the same budget, Kimi K2.6 found an average of 4.06 bugs, slightly outperforming GPT-5…
-
AI models tested on admitting ignorance, Claude hedges, Qwen3.8 Max errs confidently
A recent test evaluated 12 AI models on their ability to admit when they don't know an answer, rather than confidently providing incorrect information. The study found that Qwen3.8 Max was the most prone to confident er…
-
DeepSeek V5 to leverage Huawei Ascend chips for pre-training
DeepSeek is planning to use Huawei's Ascend chips for the pre-training of its upcoming DeepSeek V5 large model. This marks a significant shift from previous uses of Ascend chips, which were primarily for inference tasks…
-
AI video models tested for dance generation, revealing editing challenges
A review of the latest AI video generation models, including GLM-5.1, DeepSeek V4-Pro, Kimi K2.6, Qwen 3.7-max, and Minimax M3, highlights the challenges in achieving ideal results. While AI can streamline production, u…
-
Uncensored LLMs slash safety refusals but lose capabilities
A recent analysis compared three pairs of language models, evaluating their safety refusals and helpfulness. The study found that "uncensored" versions of models like Llama 3.1 8B, Qwen2.5 7B, and Mistral 7B Instruct si…
-
New DeepDumb benchmark tests AI patience against confident stupidity
A new benchmark called DeepDumb has been developed to test AI models' resilience against "confident stupidity" rather than sophisticated adversarial attacks. The benchmark uses a dataset of 4,000 intentionally nonsensic…
-
AI models forecast 2026 US midterms, show no consensus on French election
A recent analysis asked 18 AI models to forecast the outcomes of the 2026 US midterm elections and the 2027 French presidential election. The AI models provided forecasts on various aspects, including party control of t…
-
New FreeCAD Fix benchmark uses deterministic scoring for LLM evaluation
The FreeCAD Fix benchmark offers a more rigorous evaluation for LLMs by employing a deterministic geometry oracle instead of an LLM judge. This method ensures that model performance is measured by the actual output of r…
-
LLMs achieve perfect scores with new tool integration
A recent experiment revealed that Large Language Models (LLMs) like DeepSeek V4.1 Flash 1017, Kimi K3.1 Pro, and DeepSeek V4 Pro perform significantly better when a tool is available for them to interact with. In a new …
-
LLM skill files: Shell version yields similar gains to full structure
A recent retest of LLM skill files revealed that while Kimi K3.1 Pro and DeepSeek V4 Pro showed gains with a full skill file, a simplified "shell" version of the file yielded similar improvements. This suggests that for…
-
Four LLMs evaluated on personalized English learning plan generation
Four large language models, Claude Sonnet 5, GPT-6.1 Sol, DeepSeek V4 Pro, and Qwen3.8 Max, were tested on their ability to create a personalized 7-day English learning plan. When given the same prompt, the models produ…