DeepSeek Flash Pricing Swung Twice This Week, and GPT-6.1 Terra Just Showed Up

Here's what our price tracker picked up over the last seven days across Claude, GPT, Gemini, DeepSeek, and Grok.

This is the headline for practical reasons. If you're using DeepSeek V4.1 Flash in production, the input price moved twice in one week:

So input got more than doubled (from $0.02 to $0.045), while output got cut in half (from $1.00 to $0.50). The net effect depends entirely on your workload. If you're doing something input-heavy — long context retrieval, document processing, big system prompts — this week ended up more expensive for you. If you're generating a lot of output relative to input, you came out ahead.

The bigger point is that two price changes in five days on the same model is unusual. Most providers move prices slowly and announce it. Seeing it shift mid-week without fanfare is a good reminder that relying on cached cost estimates in your head (or in a spreadsheet from last month) can quietly break your budget assumptions. Worth re-checking before you commit to a contract or a long-running batch job.

We also spotted a pricing record for Kimi K3.1 Pro on Oct 27: $4.00 per million input tokens, $20.00 per million output tokens. That puts it in the same general tier as GPT-4-class models — not cheap, not the most expensive thing out there, but clearly positioned as a premium offering rather than a budget option.

We don't have a prior price snapshot for this model to compare against, so I can't tell you if this is a change or just the first time we caught it. Either way, if Kimi is on your radar, that's the current number to work with.

Four new OpenAI model IDs showed up in our tracker on Oct 27:

The batch variants are the standard pattern OpenAI uses for async/discounted batch processing. What's notable here is the "Terra" naming — it's not a suffix OpenAI has used before in models we track. Whether that's a new capability tier, a regional thing, or something else entirely, I genuinely don't know yet. We're just reporting what appeared.

Pricing for these models wasn't in the data we pulled, so I can't give you numbers. Keep an eye on it if you're evaluating GPT-6.1 options — the batch variants in particular tend to come in at a significant discount to the synchronous versions, which matters if your use case can tolerate async turnaround.

A few takeaways if you're making model decisions right now:

DeepSeek Flash users: re-run your cost math with the Oct 31 prices ($0.045 in / $0.50 out). If your workload is output-heavy, you actually got a break. If it's input-heavy, you didn't.

GPT-6.1 Terra: too early to evaluate cost-effectiveness without pricing, but it's worth knowing it exists. The batch variants will likely be the interesting ones once pricing lands.

General: this week is a good illustration of why static pricing docs go stale. If you're building something where API costs matter, you want a live source, not a blog post from three months ago.

We track prices daily across OpenAI, Anthropic, Google, DeepSeek, and xAI. Full current pricing is at llmpricewatch.com.