LLM Pricing Digest: DeepSeek V4, Gemini 3 Flash Prices, and Four New Models This Week

Weekly pricing data from LLM Price Watch — tracking Claude, GPT, Gemini, DeepSeek, and Grok.

---

This week had a bit of both: a handful of price data points surfacing on newer models, and four new model IDs appearing in our tracker. Here's what actually showed up in the data.

Price Data: What We're Seeing

Gemini 3.8 Flash

Our tracker logged pricing for gemini-3-8-flash at $0.825/M input and $4.125/M output (checked October 18). That output price is notably higher than what Google has trained us to expect from their Flash-tier models. Gemini 2.0 Flash, for comparison, has been sitting at fractions of that on output. If you're running high-output workloads — summaries, long-form generation, agentic loops that produce a lot of tokens — that output cost is worth paying attention to before you swap this in as a drop-in upgrade.

DeepSeek V4 Flash and V4 Pro

Two DeepSeek V4 variants logged pricing on October 12:

• deepseek-v4-1-flash: $0.30/M input, $1.20/M output
• deepseek-v4-pro: $1.32/M input, $3.96/M output

The Flash variant is extremely cheap. At $0.30 input and $1.20 output, it's in territory that makes it worth testing for anything where cost is the primary constraint — classification, routing, structured extraction, that kind of work. Whether it's good enough for your use case is a different question, but the price is hard to ignore.

The Pro variant at $1.32/$3.96 sits in a more competitive range. It's cheaper than GPT-4o and in the same general neighborhood as Claude Haiku 3.5 on input, though output pricing comparisons depend on your token ratios. If DeepSeek V4 Pro holds up quality-wise, it could be a legitimate contender for mid-tier tasks where you want something better than Flash but don't need to pay frontier prices.

New Models Spotted

Four new model IDs showed up in our tracker this week:

deepseek/deepseek-v4.1-flash-1017 (first seen October 18) — This looks like a dated checkpoint or updated version of the V4 Flash line. DeepSeek has a pattern of releasing timestamped variants, so this is worth watching if you're pinning to specific model versions in production.

qwen/qwen3.7-plus:batch (first seen October 17) — A batch variant of Qwen 3.7 Plus from Alibaba/Qwen. Batch endpoints typically come with meaningful price discounts in exchange for slower turnaround, so if you have async workloads this is worth checking pricing on once it's fully listed.

google/gemini-4-argon and google/gemini-4-argon:batch (both first seen October 14) — This is the most interesting new appearance. Gemini 4 Argon, including a batch variant, showing up in the API. "Argon" as a codename suggests this might be a specific tier or configuration within the Gemini 4 family rather than the flagship. No pricing data in our system yet, so we can't say much about cost positioning.

What This Means Practically

If you're evaluating models right now:

• High-volume, cost-sensitive work: DeepSeek V4 Flash at $0.30/$1.20 is worth benchmarking. Just test it against your actual tasks first.
• Mid-tier reasoning tasks: DeepSeek V4 Pro enters a crowded field, but the pricing is competitive enough to put it on your shortlist.
• Gemini users: The Gemini 3.8 Flash output price is higher than the Flash name might imply — run your token math before assuming it's cheaper than what you're using today.
• Keep an eye on Gemini 4 Argon: No pricing yet, but two variants appearing simultaneously (standard + batch) suggests Google is preparing a real release.

Prices change fast. The data above reflects what our tracker detected this week — by the time you read this, some of it may already have shifted.

---

For live pricing across all five providers, check llmpricewatch.com.