Changelog

New models, price changes and retirements on Entrim. Every announcement we email, in one place.

DeepSeek V4.1 Flash is LIVE

We said last week it would land in days, and it's here. V4.1 brings stronger performance on coding and agentic workloads, along with new capabilities over V4 Flash.

Now with vision

V4.1 Flash can process images alongside text. Send screenshots, diagrams, charts, scanned pages, or other visual content directly in the prompt for analysis, extraction, or as context for the task.


#01NEWDEEPSEEK

Start testing it today

If you're already calling our API, only the model name changes:

Model name
"deepseek-ai/DeepSeek-V4.1-Flash"

Pricing

Input / M
$0.20
Output / M
$0.60
Cached / M
$0.005

Currently in beta

V4.1 will stay in beta while we tune its performance. If you see unexpected responses, errors, or performance issues, post in our Discord. Your feedback goes straight to our engineers.


#02BENCHMARKSV4.1 vs V4

V4.1 Flash vs V4 Flash

BenchmarkV4 FlashV4.1 Flash
DeepSWE v1.154.474.2
Terminal-Bench 2.182.790.6
AutomationBench37.754.8
Codeforces rating3,2893,471

#03COMINGV4 FLASH

V4 Flash remains available for now

V4 Flash isn't deprecated yet, and we haven't set a retirement date. Part of its serving capacity has now been reallocated to V4.1, so you may notice lower throughput than you're used to. You can still expect around 100 tokens/s.

If you're currently using V4 Flash, we recommend starting your migration to V4.1 now. We'll announce the retirement date when V4 Flash enters deprecation, with advance notice before API access ends.

Try DeepSeek V4.1 Flash

#01STABLEGLM

GLM 5.3 Flash is out of beta

It's been in beta since release. It's now stable and ready for production work, and faster along the way: median time to first token is down from 2.1s to 1.2s.

Thank you for every issue you reported. When we work together while a model is in beta, we can make sure you get the best performance and quality, and get it stable as quickly as possible.


#02DEPRECATIONQWEN

Qwen 3.6 27B is retiring on October 6, 2026

When we released Qwen 3.8 27B we said you'd stay on 3.6 until it ran exactly the way it should. It does: better reasoning, stronger coding, more reliable tool calling.

Retires
Oct 6
Replacement
3.8 27B
Price change
None

What you need to do

Model APIs Swap the model identifier wherever you set it, before October 6:

Model name
Old: "Qwen/Qwen3.6-27B"
New: "Qwen/Qwen3.8-27B"

Model Plans Nothing to do on your end: we'll move your slot before October 6, same plan, same price, and your current model name keeps working. We'd still recommend swapping it to "Qwen/Qwen3.8-27B", so your setup says what it's actually running. To switch early, use your dashboard under Available models.


#03IMPROVEDINFRASTRUCTURE

We now serve 4x the tokens we could a month ago

We upgraded our capacity at the same prices, and improved how we serve the models so more of that capacity reaches your requests.


#04DELAYEDAPI KEYS

Multiple API keys got delayed

We said they'd ship last week and they didn't. What looked like a few hours of work turned out to need real changes to the platform architecture underneath, and we'd rather fix that properly than ship something we have to unpick later.

Sorry for the wait. If everything goes to plan it should take up to three weeks, and you'll hear the moment it's live.


#05COMINGDEEPSEEK

DeepSeek V4.1 goes live this week

Last checks are running now. The capacity above is what made room for it.

#01GLM

GLM 5.3 Flash is now available in beta

You can start using and testing GLM 5.3 Flash today. It will remain in beta while we gather your feedback and make improvements.

As you test the model, please share any unexpected responses, errors, performance issues, or suggestions on our Discord.

Pricing per million tokens:

Input / M
$0.10
Output / M
$0.35
Cached / M
$0.0175

We’re currently seeing a 97% cache hit rate, helping reduce input costs through cached pricing.


#02QWEN

Qwen 3.6 35B-A3B deprecation: September 28, 2026

We plan to deprecate Qwen 3.6 35B-A3B on September 28, as usage is approaching the minimum level needed for us to continue serving the model. If you currently use it, please plan to switch to another available model before September 28.

#01DEEPSEEK

A new DeepSeek V4 Flash release

Last week we upgraded DeepSeek V4 Flash to version 0731. This week we made a new release where we fixed some of the issues you reported:

  • More capacity.We added extra serving capacity, so performance is even better than before
  • Fixed strange outputs.A few of you reported odd completions — those cases should be fixed now.

Everything is already live. Your requests picked up the new release automatically, so there's nothing you need to do.


#02QWEN

Qwen 3.8 27B is finally here — grab the model-id and test it

We've released the newest Qwen 3.8 27B model on Model APIs.

It's a clear step up from the 3.6 generation — better reasoning, stronger coding, and more reliable tool calling — with the same OpenAI-compatible interface you already use.

Pricing is the same as Qwen 3.6 27B:

Input / M
$0.10
Output / M
$0.40
Cached / M
$0.04

It's brand new, and we'd love your help breaking it in. If anything looks off, just write us on Discord — it goes straight to the engineers. That's exactly how the DeepSeek fixes above happened.

How to get on it

Model APIs It's one line. Grab the name from the model library and you're done:

Model name
model="Qwen/Qwen3.8-27B"

Model Plans Nothing to do on your end: you stay on 3.6 until 3.8 runs exactly the way it should, then we flip the default. Don't want to wait? Swap "Qwen/Qwen3.6-27B" for "Qwen/Qwen3.8-27B" and you're on it — and you can switch back anytime. If anything breaks, tell us.


#03PRICING

DeepSeek is raising prices. We're keeping ours.

You may have seen the latest news: DeepSeek announced a price increase with this release. We've decided not to follow. Our prices stay exactly where they are:

Input / M
$0.09
Output / M
$0.17
Cached / M
$0.015

So you get the newer, faster model — and your bill doesn't change.

Get every update by email

Same news, sent to your inbox when it ships. Unsubscribe anytime.

All services are online

© 2026. Entrim. All Rights Reserved.

Privacy policy•Terms of service