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:batchonce 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.
Top comments (0)