DeepSeek shipped V4.1-Flash on September 10, 2026. The interesting part: they say this Flash model beats their own V4-Pro on performance, cost, speed, and total runtime. So they are routing Pro traffic to Flash.
That is good news for PressBot in two places. Your public chatbot (what visitors see) and your admin agent (what you use in wp-admin). Point both at deepseek-flash and you usually get better answers for less money. We are adding that model id to PressBot as soon as the API is real. Same habit as every other big model drop.
Better visitor chat without lighting money on fire
Public chat is where cost hurts. Lots of questions, lots of tokens. A fancy model that is slow and expensive means you either water down the bot or watch the bill climb. V4.1-Flash sits in a nicer spot: solid scores on coding and agent-style tasks, vision built in, Flash pricing. Visitors get clearer answers on products, docs, and “what is this page about?” You stretch the same budget further.
Admin work benefits too (drafts, tools, knowledge lookups). Most site owners notice the visitor side first.
We add new models when they ship
PressBot is bring-your-own-key. When DeepSeek, OpenAI, Google, or Anthropic release something that actually moves cost or quality, we update the catalog. You pick the new id in Settings. Public chat and the admin agent both use it. No waiting on a quarterly marketing cycle.
V4.1-Flash is this week’s example. There will be another one next month. That is the point.
Pro traffic is moving to Flash
Normally “Pro” means the expensive smarter model. DeepSeek’s own tests (and the early score tables around the launch) say Flash cleared that bar. Here is their plan:
- Use
deepseek-flashas the real model id - Old names
deepseek-v4-flashanddeepseek-v4-flash-vision-expstill work for now; they route to V4.1-Flash - From about 04:00 UTC on Sep 14, 2026, until V4.1-Pro exists,
deepseek-v4-proalso routes to V4.1-Flash and bills at Flash rates
If your site still selects deepseek-v4-pro, you will land on Flash anyway after that date. Better to switch to deepseek-flash yourself and run a few tests first.
A few scores worth caring about
These numbers come from DeepSeek and early launch writeups. Independent scoreboards may still be catching up, so treat them as directional:
| Suite | V4.1-Flash | Older V4-Flash |
|---|---|---|
| Terminal-Bench 2.1 | ~90.6 | ~82.7 |
| CyberGym | ~88.1 | ~76.7 |
| GPQA Diamond | ~90.9 | — |
For WordPress (visitor Q&A, admin tools, multi-step tasks), those suites matter more than a generic chat leaderboard. The jump from the previous Flash is big.
Pricing
DeepSeek also cut Flash API prices (their schedule around Sep 10, 2026 UTC). Off-peak is half of peak. Rough off-peak Flash rates:
- Cache-hit input: about $0.003 per 1M tokens
- Uncached input: about $0.15 per 1M tokens
- Output: about $0.60 per 1M tokens
Peak is about double. Check DeepSeek’s pricing page before you point heavy traffic at it. Rates change.
Compared with the old V4-Pro list prices people were quoting at launch, Flash is a different ballpark. That is why the Pro routing change matters if you were paying Pro rates for “Pro quality.”
Vision is in the base model now
You do not need a separate vision-exp endpoint for image prompts anymore. One id covers text and vision. Handy when a visitor uploads a photo or your agent looks at a screenshot.
Using it in PressBot
Bring your own DeepSeek key. Same key can drive the public chatbot and the admin agent.
- Install or update PressBot from your Freemius zip (not wordpress.org).
- Go to PressBot → Settings, choose DeepSeek, paste the key.
- Pick
deepseek-flashin the model list once the catalog update is on your site. - Save. Try a few real visitor questions and one admin task before you make it the default everywhere.
Want missions, Quiet Shield, and the full Pro tool set? That is PressBot Pro. It is separate from which DeepSeek model you select.
Test before you flip production
New model, new defaults, new ways to break. Before you switch live traffic:
- Re-run your common visitor FAQs and tool flows
- Try one image question if you use vision
- Check speed during your real peak hours, not only in a quiet test
DeepSeek is betting on Flash as the workhorse. In PressBot, choose deepseek-flash, verify it on your site, then let the public chatbot take the win.
