DEV Community: Anders Rasmussen

LLM Pricing This Week: DeepSeek Quietly Drops a Vision Model

Anders Rasmussen — Sun, 23 Aug 2026 10:02:11 +0000

LLM Pricing This Week: DeepSeek Quietly Drops a Vision Model

Weekly digest from LLM Price Watch — we track live pricing for Claude, GPT, Gemini, DeepSeek, and Grok so you don't have to.

It was a quiet week on the pricing front. No cuts, no hikes, nothing moved across the five providers we track. If you're mid-project and budgeting around current rates, your spreadsheet is still valid.

The only thing worth noting this week is a new model showing up in the wild.

New: deepseek/deepseek-v4-flash-vision-exp

On August 22nd, our tracker spotted deepseek/deepseek-v4-flash-vision-exp appearing on OpenRouter for the first time. It's sitting under DeepSeek's lineup, and the name tells you a few things:

v4 — this is positioned as their fourth-generation base
flash — expect a speed/cost optimized variant rather than a full-capability flagship
vision — multimodal, so it can handle image inputs alongside text
exp — experimental, meaning DeepSeek hasn't called this production-ready yet

The exp tag is worth paying attention to. Experimental models from any provider can change behavior, pricing, or disappear entirely without much notice. If you're evaluating this for something that needs to stay stable, treat it as a preview rather than a dependency.

That said, DeepSeek's "flash" tier has generally meant aggressive pricing in the past, and a vision-capable model in that tier is interesting. A lot of use cases — document parsing, receipt extraction, basic image captioning — don't need the heaviest multimodal model available. If DeepSeek prices this the way they've priced their other flash variants, it could be worth benchmarking against GPT-4o mini or Gemini Flash for vision tasks.

We don't have confirmed pricing locked in for this model yet since it just appeared. We'll update the tracker as that settles.

The Bigger Picture This Week

Honestly, a week with no price changes and one experimental model drop is pretty normal. The pace of model releases has been fast enough over the past year that it's easy to assume something significant happens every week, but sometimes the answer is just: nothing moved, here's what's new.

A few things I'd keep in mind heading into next week:

DeepSeek continues to ship fast. Whether or not deepseek-v4-flash-vision-exp turns into a production model worth using, the cadence from DeepSeek has been consistent. They tend to iterate in public, so experimental models often do graduate to stable relatively quickly.

Vision is getting commoditized. A year ago, image input was a premium feature. Now it's showing up in flash/lite tiers across providers. If you're paying for a heavier model primarily because you need vision and assumed cheaper options didn't have it, it's worth re-checking the landscape.

No price changes this week doesn't mean no price pressure. Several providers have cut prices in 2025, and competition hasn't slowed. Stable prices one week doesn't signal a plateau — it just means nothing happened this particular week.

That's the week. One new experimental model from DeepSeek, everything else held steady. If you want to keep an eye on pricing as it shifts — especially for DeepSeek's new addition once pricing confirms — the tracker is running daily checks.

→ llmpricewatch.com

New Models From Google, DeepSeek, and xAI This Week — LLM Price Watch Digest

Anders Rasmussen — Sun, 16 Aug 2026 10:01:33 +0000

LLM Price Watch Weekly Digest — Week of August 14, 2026

No price changes to report this week across Claude, GPT, Gemini, DeepSeek, and Grok. That's actually notable in itself — it's been a period of relative pricing stability. But it wasn't a quiet week model-wise. Four new models showed up in our tracker across three providers, and they're worth knowing about if you're evaluating what to build on.

Google: Gemini 3.7 Flash (and a Batch Variant)

The two biggest additions are google/gemini-3.7-flash and google/gemini-3.7-flash:batch, both first spotted on August 14th.

Gemini Flash has been Google's go-to for cost-efficient, lower-latency tasks — the kind of workloads where you need volume without burning through budget. The 3.7 iteration landing now suggests Google is continuing to iterate on that line in parallel with their heavier models.

The batch variant is the one I'd pay attention to if you're running anything offline — document processing, evals, bulk classification, that sort of thing. Batch endpoints typically come with a meaningful price discount in exchange for higher latency, so if your use case doesn't need a real-time response, it's worth checking the pricing before defaulting to the standard endpoint. We'll have the confirmed pricing posted on LLM Price Watch as soon as it's stable in the tracker.

DeepSeek: v4 Pro 0813

deepseek/deepseek-v4-pro-0813 showed up on August 13th. The 0813 suffix is a date stamp — a common convention DeepSeek uses to version checkpoint releases. This appears to be a refreshed checkpoint of DeepSeek v4 Pro rather than an entirely new architecture.

DeepSeek has been one of the more interesting providers to watch from a price-to-performance standpoint over the past year. Their models have consistently undercut comparable Western models on cost. If you've been using an earlier v4 Pro checkpoint, it's worth running a quick eval against this one to see if quality has shifted. Sometimes these point releases are minor; sometimes they're not.

xAI: Grok 4.6

x-ai/grok-4.6 also appeared on August 13th. xAI has been incrementing Grok's version numbers at a reasonable clip. Grok 4.6 sits between major releases, so this reads like a refinement update — likely improvements to instruction following or reasoning rather than a fundamentally different model.

Grok has carved out a niche for users who want strong general-purpose performance and are already in the xAI ecosystem. Pricing on this one we'll confirm as the data firms up.

What This Week Means Practically

If you're in the middle of a model selection decision right now, here's how I'd think about these additions:

High-volume, latency-tolerant workloads: Look at gemini-3.7-flash:batch once pricing is confirmed. Batch endpoints are consistently underused by developers who could benefit from them.
Cost-sensitive inference: Keep an eye on the DeepSeek v4 Pro 0813 pricing. DeepSeek tends to be aggressive here and a checkpoint update sometimes comes with a price adjustment.
General-purpose API work: Grok 4.6 is worth a quick benchmark if you're already evaluating xAI models.

The lack of price changes this week means your existing cost projections are still valid — no surprises there. But new model versions mean your performance baselines might shift if providers quietly improve quality at the same price point.

I track all of this daily at LLM Price Watch — live pricing for Claude, GPT, Gemini, DeepSeek, and Grok in one place, with alerts when something changes. Worth bookmarking if you're making cost-sensitive model decisions.

Two New Models Just Dropped: DeepSeek V4 Pro and Grok 4.6 — LLM Pricing Digest

Anders Rasmussen — Thu, 13 Aug 2026 14:47:08 +0000

Two New Models Just Dropped: DeepSeek V4 Pro and Grok 4.6 — LLM Pricing Digest

Weekly roundup from LLM Price Watch — we track live pricing across Claude, GPT, Gemini, DeepSeek, and Grok so you don't have to.

No price cuts or hikes to report this week across our tracked providers. What we did catch: two new model IDs appearing in the wild on August 13th — one from DeepSeek and one from xAI. Here's what we know so far.

DeepSeek V4 Pro (0813)

Model ID: deepseek/deepseek-v4-pro-0813

DeepSeek continues their habit of date-stamping model releases, which is actually a useful practice — it makes it easy to tell at a glance which checkpoint you're running. The "V4 Pro" naming suggests this sits above their previous V3 line, though we don't yet have benchmark comparisons or confirmed context window specs to share. The 0813 suffix matches the date it first appeared in our tracker.

DeepSeek has consistently been one of the more cost-competitive options in our index, so if V4 Pro follows that pattern, it's worth watching. We'll update pricing on the site as soon as it's confirmed.

Practical take: If you're currently running DeepSeek V3 in production and cost-efficiency is a priority, keep an eye on how V4 Pro prices in. DeepSeek's previous generational jumps have held the line on price while improving capability — but don't assume that until the numbers are public.

Grok 4.6

Model ID: x-ai/grok-4.6

xAI is pushing another point release with Grok 4.6, spotted the same day as the DeepSeek drop. Grok 4 was already in our tracker, so this is an incremental update rather than a major new generation. Point releases from xAI have sometimes come with quiet capability improvements or adjusted rate limits, so it's worth checking if you're an active Grok user.

Pricing for 4.6 isn't confirmed in our system yet. Grok 4 has sat at a premium compared to some of the DeepSeek options, so if 4.6 comes in at the same price point, the question is whether the incremental improvements justify staying on that tier versus alternatives.

Practical take: If you're already using Grok 4 via API, test 4.6 on your actual use cases before migrating. Point releases don't always move the needle on the tasks that matter most to your specific workload.

The Bigger Picture This Week

Two new models, zero price changes. That's actually a notable signal in itself. We're in a period where the model release cadence remains fast, but the pricing floor seems to have stabilized — at least for this week. The race to the bottom on token pricing that defined much of early 2025 appears to have leveled off, at least temporarily.

For anyone building on top of these APIs right now, the practical advice is the same as always: don't hard-code model names if you can avoid it, because the deepseek-v4-pro-0813 you integrate today may be superseded by a -0901 variant before your next sprint is done. Abstract your model selection layer where possible.

We'll be watching both of these new models closely over the coming days as pricing details get confirmed and early benchmarks surface.

I track these changes daily at *llmpricewatch.com** — live pricing for Claude, GPT, Gemini, DeepSeek, and Grok, updated automatically. If you want alerts when something changes, the site has you covered.*

I built a pricing API for LLMs — then realized the real users might not be human

Anders Rasmussen — Sat, 08 Aug 2026 06:02:20 +0000

Klikk her: LLM Price Watch started as a simple problem: comparing per-token pricing across Claude, GPT, Gemini, DeepSeek, and Grok meant opening five pricing pages and doing the math by hand every time a new model dropped. So I built a calculator. Then I built an API behind it. Then I noticed something about who was actually going to call that API.

The obvious version

The first version of the API was exactly what you'd expect:

GET /v1/models — every tracked model with current pricing
GET /v1/models/:id — a single model
GET /v1/calculate?model=X&input_tokens=N&output_tokens=N — cost for a specific call

Straightforward. A human developer hits /calculate, gets a number, builds their cost estimate into a dashboard somewhere. Done.

The part that changed the design

The actual differentiator turned out to be a fourth endpoint: GET /v1/recommend?use_case=X. Instead of just returning prices, it returns a recommendation — which model fits a given use case (long-document summarization, high-volume classification, coding assistance, customer support) based on both price and the editorial analysis already written for the comparison pages on the site.

Once that endpoint existed, the actual audience for this API stopped being "a developer building a cost dashboard" and started including something else: AI agents doing their own tool selection at runtime. An agent framework deciding which model to route a task to doesn't want to read a blog post — it wants a structured answer to "given this use case, what should I use, and what will it cost me." That's a tool call, not a page view.

That reframing changed a few concrete decisions:

CORS is wide open on purpose. This isn't an API with a dashboard in front of it — it's meant to be called directly from wherever the calling code lives, including client-side agent code.
No API key required (for now). Every bit of friction between "an agent wants this data" and "an agent gets this data" is friction against the actual use case. A paid tier with rate limits is the natural future step once real usage justifies it, but gating from day one would have defeated the point.
/recommend returns reasoning, not just a model name. An agent — or the person who built it — needs to know why, not just what, or the recommendation is a black box nobody trusts enough to actually wire into a decision.

The unglamorous half of this

None of that matters if the numbers are wrong. Pricing data for five providers was verified directly against each provider's own official pricing page, not pulled from a third-party aggregator — aggregators lag, and stale pricing data is worse than no data for something meant to inform actual spend decisions. Updates are still a manual snapshot for now; an auto-refreshing worker is the obvious next step once there's enough usage to justify the engineering time.

The part I didn't expect

The API ended up feeding the other site I run, StackIndex AI — its cost calculator page calls this API live, gets a model recommendation plus a cost estimate, and returns it inline. Two separate sites, same backend, CORS'd across domains, tested end-to-end. I didn't plan the API to be reusable infrastructure when I built it — it just turned out that "structured, agent-callable, reasoning-included" is a useful shape for more than one problem.

If you're building something similar: design the response for a reader who can't ask a follow-up question. That constraint does more for API design than almost anything else.