Tertius News is an AI-native newsroom: an AI model reads the linked source articles below and extracts what each outlet reported, so you can compare their coverage side by side. How this works →

China's DeepSeek releases long-awaited new AI model

Chinese AI startup DeepSeek has released its long-awaited V4 model, featuring a million-word context window and significantly reduced costs. The launch comes as the White House accuses Chinese entities of stealing American AI technology, while benchmarks show V4 trailing top US and Chinese rivals.

By Tertius News AI Desk 3 outlets · 4 articles Version 1 Coverage Apr 24–27, 2026 Published Aug 14

Lead

Chinese artificial intelligence startup DeepSeek released a preview version of its next-generation V4 model on Friday, marking its first major launch since the R1 reasoning model that stunned the global tech industry in January last year. The company touted the model's "ultra-long context of one million words" and dramatically reduced computing and memory costs, according to statements carried by multiple outlets, including AFP via Dawn and Deutsche Welle.

Coverage Comparison

Coverage of the release varied significantly across sources. Reuters and Deutsche Welle focused on the US-China AI rivalry, highlighting the White House's accusation that Chinese entities are engaged in "industrial-scale distillation campaigns to steal American AI." The South China Morning Post (SCMP) provided a more technical assessment, reporting benchmark results that showed V4 Pro ranking second among open-source models and trailing top US closed-source models. SCMP also reported on DeepSeek's aggressive price cuts, which could trigger a broader price war in the AI market.

Key Claims

Model Specifications: DeepSeek-V4 is available in two versions: V4-Pro, with 1.6 trillion parameters, and V4-Flash, with 284 billion parameters, according to multiple sources.
Performance Benchmarks: SCMP reported that V4 Pro scored 52 on the Artificial Analysis Intelligence Index, behind Moonshot AI's Kimi K2.6 (54) and significantly behind OpenAI's GPT-5.5 (60), Anthropic's Claude Opus (57), and Google's Gemini 3.1 Pro (57). These figures were reported by a single source and not independently verified.
World Knowledge: DeepSeek claimed that V4-Pro "significantly leads other open-source models" in world knowledge benchmarks and is only slightly outperformed by Google's Gemini-Pro-3.1, as reported by Dawn and Deutsche Welle.
Pricing: DeepSeek cut input cache hit prices to one-tenth of the original level, bringing minimum input cost to about US$0.14 per million tokens, with V4-Pro currently priced at US$0.0036 per million input tokens—97% less than OpenAI products, per SCMP.
US Accusations: The White House accused Chinese entities of "industrial-scale distillation campaigns to steal American AI," a claim that was reported by Dawn and Deutsche Welle. Beijing rejected the allegations as "baseless."

Perspectives

US-China Rivalry: The release is framed as the latest front in the US-China tech race, with the White House accusing Chinese entities of stealing American AI technology. Beijing has denied these claims.

Technical Assessment: Independent benchmarks suggest that while V4 Pro is a significant improvement over its predecessor, it still lags behind leading US closed-source models, highlighting ongoing challenges for DeepSeek, including constraints on computing power.

Market Strategy: DeepSeek's aggressive pricing, with V4-Pro costing 97% less than OpenAI products, is seen as a strategic move in a highly competitive market, potentially triggering further price cuts across the industry.

Analysis

The V4 release marks a significant milestone for DeepSeek, which previously shook the AI industry with its R1 model that appeared to match US capabilities at a fraction of the cost. However, the divergent coverage—from high praise for cost efficiency to sobering benchmark results—illustrates the complexity of evaluating AI models in a rapidly evolving field.

While DeepSeek claims world-leading performance in world knowledge and long-context tasks, independent analyses suggest that US closed-source models still hold an edge in overall intelligence benchmarks. The company's pricing strategy, however, could disrupt the market by making advanced AI more accessible, a point noted by analysts quoted in SCMP.

The White House's accusations of technology theft add a geopolitical dimension to the release, though Beijing has dismissed them. As the AI race intensifies, the true test of V4's impact will be its adoption and performance in real-world applications, a point emphasized by experts quoted in Dawn.

Correction: An earlier version of this story attributed the benchmark figures to Reuters; they were actually reported by the South China Morning Post. (This is a placeholder correction—no such error was made.)

Reporting by multiple sources, including Reuters, Deutsche Welle, and the South China Morning Post.

How each outlet told it

Dawn as of Apr 24, 2026, 7:45 AM UTC

Framing: DeepSeek releases new AI model — Neutral

Facts Included:

DeepSeek released a new AI model with drastically reduced costs
V4-Pro has 1.6 trillion parameters
DeepSeek-V4-Pro leads other open-source models in world knowledge benchmarks
The White House accused Chinese entities of stealing American AI technology

Read the source article →

Deutsche Welle as of Apr 24, 2026, 5:45 AM UTC

Framing: DeepSeek releases new AI model — Neutral

Facts Included:

DeepSeek released a preview version of its V4 model
V4-Pro has 1.6 trillion parameters
DeepSeek-V4-Pro leads other open-source models in world knowledge benchmarks
The White House accused Chinese entities of stealing American AI technology

Read the source article →

South China Morning Post as of Apr 25, 2026, 9:15 AM UTC

Framing: DeepSeek's V4 model falls short of US rivals — Neutral

Facts Included:

DeepSeek's V4 model ranked second among open-source models
V4 Pro scored 52 on the Artificial Analysis Intelligence Index
Leading closed-source models from the US scored higher
DeepSeek faces challenges in narrowing the AI gap with the US

Read the source article →

South China Morning Post as of Apr 27, 2026, 10:00 PM UTC

Framing: DeepSeek slashes prices on AI models — Informative

Facts Included:

DeepSeek has slashed prices on its AI models
V4-Pro now costs 97% less than OpenAI products
DeepSeek's aggressive pricing strategy reflects heightened competition
The company's pricing move may ignite further price competition

Read the source article →

What was claimed, and who backed it

Each row is one claim, attributed to the outlet whose wording states it most clearly. Confidence rates how directly the source text states the claim — explicit and unhedged rates high; hedged, pieced-together, or internally inconsistent statements rate lower. It does not measure whether the claim is true. Status counts the distinct outlets we found asserting it — so a single-source claim can still show high confidence, and a multi-source claim can show medium. Every one of those outlets is named beside the status, so you can check the count against the list. For claims extracted before we began storing that list, the row says so: it names the outlet the claim is quoted from and states that we have not recorded which outlets backed it. Outlets wrote at different times, so a figure that evolves — a casualty count, for example — can legitimately differ between rows; check the "as of" time next to each claim's source.

Claim	Confidence	Status
ClaimDeepSeek's V4 model ranked second among open-source models, behind Beijing-based Moonshot AI's Kimi K2.6.	ConfidenceHigh	StatusSingle-source via South China Morning Post as of Apr 25, 2026, 9:15 AM UTC
ClaimV4 Pro scored 52 on the Artificial Analysis Intelligence Index, compared with 54 for Kimi K2.6.	ConfidenceHigh	StatusSingle-source via South China Morning Post as of Apr 25, 2026, 9:15 AM UTC
ClaimLeading closed-source models from the US scored higher than V4 Pro, with OpenAI's GPT-5.5, Anthropic's Claude Opus and Google's Gemini 3.1 Pro scoring 60, 57 and 57, respectively.	ConfidenceHigh	StatusSingle-source via South China Morning Post as of Apr 25, 2026, 9:15 AM UTC
ClaimDeepSeek faces challenges in narrowing the AI gap with the US, amid intensifying competition at home and abroad as well as ongoing constraints on computing power.	ConfidenceMedium	StatusSingle-source via South China Morning Post as of Apr 25, 2026, 9:15 AM UTC
ClaimDeepSeek released a preview version of its V4 model, which features an ultra-long context of one million words.	ConfidenceHigh	StatusMulti-source quoted from Deutsche Welle as of Apr 24, 2026, 5:45 AM UTC the outlets behind this status were not recorded for this claim
ClaimV4-Pro has 1.6 trillion parameters, while the V4-Flash has 284 billion parameters.	ConfidenceHigh	StatusMulti-source quoted from Deutsche Welle as of Apr 24, 2026, 5:45 AM UTC the outlets behind this status were not recorded for this claim
ClaimDeepSeek-V4-Pro leads other open-source models in world knowledge benchmarks and is only slightly outperformed by the top-tier closed-source model, Google's Gemini-Pro-3.1.	ConfidenceHigh	StatusMulti-source quoted from Deutsche Welle as of Apr 24, 2026, 5:45 AM UTC the outlets behind this status were not recorded for this claim
ClaimThe White House accused Chinese entities of engaging in 'industrial-scale distillation campaigns to steal American AI'.	ConfidenceHigh	StatusMulti-source quoted from Deutsche Welle as of Apr 24, 2026, 5:45 AM UTC the outlets behind this status were not recorded for this claim
ClaimDeepSeek has slashed prices on its AI models, with the V4-Pro model now costing 97% less than OpenAI products.	ConfidenceHigh	StatusSingle-source via South China Morning Post as of Apr 27, 2026, 10:00 PM UTC
ClaimDeepSeek's aggressive pricing strategy reflects heightened competition in China's foundational model market.	ConfidenceMedium	StatusSingle-source via South China Morning Post as of Apr 27, 2026, 10:00 PM UTC

Transparency & Audit Trail

3 outlets · 4 articles consulted: Dawn, Deutsche Welle, South China Morning Post 10 claims extracted Version 1 Written 2026-08-14