share big model of deepseekv4

The DeepSeek-V4 series abandons the single model specification and offers two versions for different scenarios: the Pro version with ultimate performance and the Flash version with ultimate cost-effectiveness. The total parameter count reaches 1.6 trillion, with the activated parameter count reaching 49 billion, and the context length is 1 million Tokens. Its core advantages lie in deep reasoning, complex code generation, and Agent tasks.

Three key features:

1. Super-long context: With the new hybrid attention mechanism (CSA+HCA) and DSA sparse attention, V4 reduces the computing power demand to 27% of the previous generation when processing a million contexts, and the KV cache to 10%.
2. Agent capabilities leap from "chatting" to "working": The core of V4 lies in its powerful agent capabilities. It is no longer just answering questions but can independently plan tasks, invoke tools, correct errors, and complete deliveries.

3. Ultimate cost-effectiveness: DeepSeek-V4 continues its inclusive pricing strategy, reducing the calling cost of top models to an extremely low level.

4. It outperformed all open-source models in evaluations such as LiveCodeBench and HMMT, and was on par with top-tier closed-source models.

5. It scored 84.4 in the Chinese-SimpleQA evaluation, significantly outperforming GPT-5.5 and Claude.

API Access (Recommended for Developers)
Visit https://teniuapi.online, register and log in, then create a token in the "Token Management" menu.
{
"env": {
"ANTHROPIC_BASE_URL": "https://teniuapi.online",
"ANTHROPIC_AUTH_TOKEN": "Your API Access Key",
"API_TIMEOUT_MS": "3000000",
"CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC": "1",
"ANTHROPIC_MODEL": "deepseek-v4-flash",
"ANTHROPIC_SMALL_FAST_MODEL": "deepseek-v4-flash",
"ANTHROPIC_DEFAULT_SONNET_MODEL": "deepseek-v4-flash",
"ANTHROPIC_DEFAULT_OPUS_MODEL": "deepseek-v4-flash",
"ANTHROPIC_DEFAULT_HAIKU_MODEL": "deepseek-v4-flash"
},
"skipDangerousModePermissionPrompt": true
}