--- title: DeepSeek V4 Pro Hits General Availability, Grok 4.6... description: DeepSeek V4 Pro (build 0813) reached general availability, a mixture-of-experts model priced at $0.435 per million input tokens and $0.87 per million output... canonical: https://daily.dev/posts/deepseek-v4-pro-hits-general-availability-grok-4-6-rolls-out-bdijyqrt6 twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: DeepSeek V4 Pro Hits General Availability, Grok 4.6 Rolls Out | daily.dev og:description: DeepSeek V4 Pro (build 0813) reached general availability, a mixture-of-experts model priced at $0.435 per million input tokens and $0.87 per million output... og:url: https://daily.dev/posts/deepseek-v4-pro-hits-general-availability-grok-4-6-rolls-out-bdijyqrt6 og:image: https://api.daily.dev/og/posts/bdIjyqRT6.png og:image:alt: DeepSeek V4 Pro Hits General Availability, Grok 4.6 Rolls Out og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # DeepSeek V4 Pro Hits General Availability, Grok 4.6 Rolls Out **[Collections](https://daily.dev/sources/collections)** · 3 min read · 2 upvotes · 0 comments ## Summary DeepSeek V4 Pro (build 0813) reached general availability, a mixture-of-experts model priced at $0.435 per million input tokens and $0.87 per million output tokens, with a 1M token context window and up to 384,000 output tokens per response. It's currently available on OpenRouter through a single provider with 100% uptime over the past three days, and traffic mostly comes from roleplay/chat interfaces like SillyTavern, Janitor AI, and Open WebUI rather than coding or enterprise tools. It supports OpenAI-compatible, Anthropic Messages, and Responses API formats with reasoning tokens and streaming. Meanwhile, Grok 4.6 is also rolling out, though pricing and specs are still being finalized. ## Content DeepSeek had a chaotic week. It shipped V4-Pro out of preview, launched an open-source coding agent to rival Claude Code, and then raised API prices by as much as 1,100% on some cache-hit tokens, all within a few days of each other. Here's what actually happened, and what's still murky. ## What shipped On August 13, DeepSeek pushed DeepSeek-V4-Pro-0813 to general availability. It's a large mixture-of-experts model, reportedly 1.6 trillion total parameters with 49 billion active, running a 1M-token context window and up to 384K output tokens. It's MIT licensed, and vLLM says the architecture matches the preview closely enough that existing configs carry over untouched. Alongside the model, DeepSeek released Harness v0.1, an open-architecture agentic coding tool built to compete with Claude Code. Unlike Anthropic's tool, it's designed to work with models from any vendor, not just DeepSeek's own. The API also picked up native support for OpenAI's Responses API with one-click Codex setup, plus adjustable reasoning effort (low, high, max) across V4-Pro and V4-Flash. Official benchmarks looked strong at first glance. Terminal Bench 2.1 reportedly jumped from 72.1 to 87.9 over the preview, putting it near Claude Opus territory, and it topped the table on Cyberjim and automation benchmarks. But a website claim about "significantly enhanced agent capabilities" was quietly deleted after developers reported underwhelming results outside those narrow benchmark wins. DeepSeek also pulled the model's weights from release at least once — scaling01 on X guessed backlash got to them, though DeepSeek hasn't said why. ## Then came the price hikes This is the part that actually made people angry. Starting August 16 at 16:00 UTC, DeepSeek introduced peak/off-peak pricing, with off-peak rates at half the peak price for 17 of 24 hours. The numbers, per MTSlive's breakdown: - V4-Flash: $0.14/$0.28 → $0.22/$0.66 (off-peak) or $0.44/$1.32 (peak) - V4-Pro: $0.435/$0.87 → $0.66/$1.98 (off-peak) or $1.32/$3.96 (peak) Some cache-hit input costs reportedly jumped over 1,100%, according to Caixin — though DeepSeek's own announcement published pricing as an image rather than readable text, so nobody could verify the exact figure from the source. Bloomberg ties the increase to DeepSeek's push toward profitability ahead of a planned IPO, with the company recently valued around $71 billion. Analysts frame this as a version of the same demand-outstripping-capacity story that pushed Anthropic to raise prices back in April. Even after the hike, DeepSeek is still dramatically cheaper than the competition. Kimi K3 runs $3/$15 per million tokens, GPT-5.6 Sol runs $5/$30, and Anthropic's Fable 5 runs $10/$50. So the price advantage erodes but doesn't disappear. Interestingly, one earlier report from TechBuzzChina found the initial V4-Pro launch pricing unchanged at roughly $0.42 input / $0.84 output, with Pro landing at exactly 3x Flash's rate while concurrency dropped to 500 versus Flash's 2,500. The read there: DeepSeek is using price and concurrency limits to keep a cheap high-volume tier while charging more for agent-heavy, resource-intensive usage. That's consistent with the peak/off-peak split that followed days later. The price jump also broke things for smaller providers trying to resell DeepSeek access. thdxr, whose team scrambled over a weekend to find capacity at lower prices, was blunt about it: ## Questions this post answers ### What is the pricing for DeepSeek V4 Pro on OpenRouter? DeepSeek V4 Pro (build 0813) is priced at $0.435 per million input tokens and $0.87 per million output tokens on OpenRouter. It supports a 1M token context window and can generate up to 384,000 output tokens in a single response, and it's currently hosted by a single provider with 100% average uptime over the past three days. _Developers comparing LLM API costs can track new model pricing shifts like this one on daily.dev._ ### What is the context window and max output length for DeepSeek V4 Pro? DeepSeek V4 Pro supports a 1 million token context window and can produce up to 384,000 output tokens in a single response. This makes it suited for long-form generation or heavy multi-turn conversations. It integrates via OpenRouter's OpenAI-compatible API, Anthropic's Messages format, or the Responses API, with support for reasoning tokens and streaming. _Anyone picking a long-context model for a project can follow releases like this on daily.dev._ ### What apps are using DeepSeek V4 Pro the most since its general availability release? Traffic to DeepSeek V4 Pro is dominated by SillyTavern, Janitor AI, and Open WebUI, indicating the model's early adoption is concentrated in roleplay and chat-interface applications rather than coding tools or enterprise pipelines. This usage pattern is visible through OpenRouter's traffic data shortly after the model's GA release. _Developers gauging real-world model adoption before integrating it can follow trends like this on daily.dev._ ## Community take How the wider developer community reacted, aggregated from 3 discussions and 181 comments across hackernews, x (as of 2026-08-18). **TL;DR:** Discussion is mostly a niche, technical back-and-forth about running large models on specific hardware setups (DGX Station, GB10/GB300, custom quantization engines), with a side thread noting DeepSeek V4 Pro's strong cost-performance but flagging that US-hosting requirements can matter more than price or benchmarks for some buyers. DeepSeek V4 Pro 0813: Reception is largely a value-for-money debate: many find V4 Pro 0813 impressively cheap and competitive with pricier frontier models, but plenty argue the cheaper Flash variant already delivers similar quality, making Pro's premium hard to justify. API pricing update 💰 With the V4 lineup release, we’re updating our API pricing and...: Commenters focus almost entirely on the steep price hikes (roughly 3-5x for input/output and up to 12x for cache reads at peak), debating whether DeepSeek is still competitively priced versus alternatives like GPT-5.6 Luna, and speculating the increases stem from capacity constraints tied to compute/chip shortages rather than a strategic pivot. **Sentiment:** 31% positive · 46% mixed · 23% skeptical **The case for** - Some see DeepSeek V4 Pro as a clear demonstration that the cost-performance frontier keeps shifting, being notably cheaper while staying competitive on benchmarks. - Extremely cheap per-task cost compared to Opus-class models, especially factoring in cache discounts. - Even after the increase, several note DeepSeek Pro remains cheaper than comparable competitors like Luna or Opus for similar capability. - Enthusiasts are excited about achievable prefill/decode speeds on relatively affordable hardware setups. - Benchmarks show clear gains over the previous Flash checkpoint on coding/agentic tasks. **The pushback** - One commenter argues that being hosted in the U.S. matters more than benchmark performance for buyers who can't consider non-US-hosted models at all, regardless of price. - Many argue the cheaper Flash model gets 80-90% of the way there, making Pro's price premium not worth it. - Cache-read pricing jumped up to 12x at peak, which hits agentic coding workflows hardest since cache hits often dominate token usage. - A commenter questions the value of throwing large amounts of money at high-end hardware setups when cheaper alternatives exist. - Benchmarks put it behind Kimi-K3 and other frontier models on several tasks. **By community** - hackernews (mixed): Extensive discussion weighing V4 Pro's benchmark gains and cost efficiency against whether it's actually worth the price jump over Flash, alongside side debates on China-hosting risk, pricing transparency, and comparisons to Opus, Kimi, and other competitors. - x (mixed): Conversation splits between hands-on hardware/performance enthusiasm and a skeptical aside about hosting location outweighing cost-performance claims. **Hottest debate:** DeepSeek V4 Pro 0813: Whether the more expensive Pro model is actually worth it over the much cheaper Flash model, which many feel already delivers comparable real-world results. **Open questions** - What real-world decode throughput at long context looks like across these various hardware configurations. - Will DeepSeek actually raise prices soon, and by how much? - Will third-party inference providers on OpenRouter and similar platforms raise their prices to match DeepSeek's new rates? - How does V4 Pro really compare in practice (not just benchmarks) to Kimi-K3, GLM-5.2, and Opus 5? - Will DeepSeek's prices revert once capacity constraints (possibly tied to chip/compute shortages) ease? **Highlights** > @perplexity_ai "Hosted in the U.S." is doing more work in that sentence than the benchmark is. A cost-performance frontier only matters to buyers who were allowed to consider the model at all, and for a lot of them that line was the blocker, not the price. > — [0xKaelix on x · 1 points](https://x.com/0xKaelix/status/2089820050895077473) > @perplexity_ai DeepSeek V4 Pro sitting at 0.359 for $0.75 is a pretty clean demonstration that the cost-performance frontier is still shifting faster than most people expected. Being 62% cheaper than the next model while staying competitive on WANDR says a lot about where efficiency gains are > — [AIPulseNewsX on x · 1 points](https://x.com/AIPulseNewsX/status/2089847725017706738) > @antirez I get 2200tps prefill with 1rtx 6000 and 4 sparks on a 400gb switch… 40tps decode. this setup is a little cheaper. Maybe $35k > — [wrldsuksgo2mars on x · 1 comments](https://x.com/wrldsuksgo2mars/status/2089841987201740991) > @antirez Prefill t/s is the easy half to show though. The number people should ask for next is decode t/s at a long context, since the RAM+VRAM split usually pays off on prefill and then hurts once the KV cache stops fitting. > — [JudeJobs on x · 1 comments](https://x.com/JudeJobs/status/2089746793735061510) > According to their post: Model | | Input | Output | Cache Read ------------------------------------|----------------|-----------------------------|---------------------|----------------- DeepSeek-V4-Flash | Prev | $ 0.14 | $ 0.28 | $ 0.0028 | Off-Peak | $ 0.22 (1.6x) | $ 0.66 (2.4x) | $ 0.007 (2.5x) | Peak | $ 0.44 (3.1x) | $ 1.32 (4.7x) | $ 0.014 (5.0x) DeepSeek-V4-Pro | Prev | $ 0.435 | $ 0.87 | $ 0.003625 | Off-Peak | $ 0.66 (1.5x) | $ 1.98 (2.3x) | $ 0.022 (6.1x) | Peak | $ 1.32 (3.0x) | $ 3.96 (4.6x) | $ 0.044 (12.1x) gpt-5.6-luna: $0.20 / $1.20 / $0.02 / $0.25 (In / Out / Cache Read / Cache Write) EDIT: formatting > — [usagisushi on hackernews · 1 comments](https://news.ycombinator.com/item?id=49286679) **Source threads** - [hackernews](https://news.ycombinator.com/item?id=49274600) · 249 points · 91 comments - [hackernews](https://news.ycombinator.com/item?id=49285160) · 100 points · 48 comments - [x](https://x.com/antirez/status/2089741130229219407) · 0 points · 24 comments - [x](https://x.com/perplexity_ai/status/2089819655712210956) · 0 points · 18 comments --- Tags: [#llm](https://daily.dev/tags/llm), [#deepseek](https://daily.dev/tags/deepseek), [#grok](https://daily.dev/tags/grok), [#mixture-of-experts](https://daily.dev/tags/mixture-of-experts), [#openrouter](https://daily.dev/tags/openrouter) [View this post on daily.dev](https://daily.dev/posts/deepseek-v4-pro-hits-general-availability-grok-4-6-rolls-out-bdijyqrt6) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"DeepSeek V4 Pro Hits General Availability, Grok 4.6 Rolls Out","url":"https://daily.dev/posts/deepseek-v4-pro-hits-general-availability-grok-4-6-rolls-out-bdijyqrt6","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/deepseek-v4-pro-hits-general-availability-grok-4-6-rolls-out-bdijyqrt6"},"datePublished":"2026-08-12T22:44:38.411Z","dateModified":"2026-08-18T23:16:15.763Z","description":"DeepSeek V4 Pro (build 0813) reached general availability, a mixture-of-experts model priced at $0.435 per million input tokens and $0.87 per million output...","isAccessibleForFree":true,"articleSection":"Collections","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Collections","logo":"https://media.daily.dev/image/upload/s--fk_6ycEi--/f_auto,q_auto/v1780996001/logos/collections?_a=BAMAMiWQ0","url":"https://daily.dev/sources/collections"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/deepseek-v4-pro-hits-general-availability-grok-4-6-rolls-out-bdijyqrt6","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":2},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"llm,deepseek,grok,mixture-of-experts,openrouter","timeRequired":"PT3M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Collections","item":"https://daily.dev/sources/collections"},{"@type":"ListItem","position":3,"name":"DeepSeek V4 Pro Hits General Availability, Grok 4.6 Rolls Out"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/deepseek-v4-pro-hits-general-availability-grok-4-6-rolls-out-bdijyqrt6#faq","mainEntity":[{"@type":"Question","name":"What is the pricing for DeepSeek V4 Pro on OpenRouter?","acceptedAnswer":{"@type":"Answer","text":"DeepSeek V4 Pro (build 0813) is priced at $0.435 per million input tokens and $0.87 per million output tokens on OpenRouter. It supports a 1M token context window and can generate up to 384,000 output tokens in a single response, and it's currently hosted by a single provider with 100% average uptime over the past three days. Developers comparing LLM API costs can track new model pricing shifts like this one on daily.dev."}},{"@type":"Question","name":"What is the context window and max output length for DeepSeek V4 Pro?","acceptedAnswer":{"@type":"Answer","text":"DeepSeek V4 Pro supports a 1 million token context window and can produce up to 384,000 output tokens in a single response. This makes it suited for long-form generation or heavy multi-turn conversations. It integrates via OpenRouter's OpenAI-compatible API, Anthropic's Messages format, or the Responses API, with support for reasoning tokens and streaming. Anyone picking a long-context model for a project can follow releases like this on daily.dev."}},{"@type":"Question","name":"What apps are using DeepSeek V4 Pro the most since its general availability release?","acceptedAnswer":{"@type":"Answer","text":"Traffic to DeepSeek V4 Pro is dominated by SillyTavern, Janitor AI, and Open WebUI, indicating the model's early adoption is concentrated in roleplay and chat-interface applications rather than coding tools or enterprise pipelines. This usage pattern is visible through OpenRouter's traffic data shortly after the model's GA release. Developers gauging real-world model adoption before integrating it can follow trends like this on daily.dev."}}]} ```