Tool RPGM DazedMTLTool - A tool that provides quality MTL translations using ChatGPT

[jaden_yuki]

gemini 3 flash is out

 

[HikariSomnium]

gemini 3 flash is out

Yeah, I’ve already managed to try it out a bit. The translation quality is roughly on par with Gemini 3 Pro, but that’s judging by the small snippets I use for testing. Of course, I still need to check the quality on at least medium-sized translation projects.



Honestly, it’s time to move away from OpenAI models entirely (not sure if to Gemini or DeepSeek yet). The current trends are just awful. I got banned a few days ago without any warning or notification. It just kicked me out of my account, and when I try to log in, it says - 'Oops! You do not have an account because it has been deleted or deactivated'.

As far as I understand, they launched new content analysis algorithms and didn't like what I was translating via the API (even though it was just regular hentai manga, nothing extreme). In the end, I lost all the funds I had deposited for the API lol

 

[jaden_yuki]

Yeah, I’ve already managed to try it out a bit. The translation quality is roughly on par with Gemini 3 Pro, but that’s judging by the small snippets I use for testing. Of course, I still need to check the quality on at least medium-sized translation projects.



Honestly, it’s time to move away from OpenAI models entirely (not sure if to Gemini or DeepSeek yet). The current trends are just awful. I got banned a few days ago without any warning or notification. It just kicked me out of my account, and when I try to log in, it says - 'Oops! You do not have an account because it has been deleted or deactivated'.

As far as I understand, they launched new content analysis algorithms and didn't like what I was translating via the API (even though it was just regular hentai manga, nothing extreme). In the end, I lost all the funds I had deposited for the API lol

let's see how the rate limit will be. I don't know if google will be more permissive for long (regarding NSFW), and I didn't like the new DeepSeek 3.2, the new Kimi K2 is still undefeated (on my test)

 

[HikariSomnium]

let's see how the rate limit will be. I don't know if google will be more permissive for long (regarding NSFW), and I didn't like the new DeepSeek 3.2, the new Kimi K2 is still undefeated (on my test)

For now, the free tier is limited to 20 requests per day and 5 requests per minute. Could you tell me which exact Kimi version gives good results (is it Kimi-K2 0905 or Kimi-K2-Thinking)? do you use it via the OpenRouter API or directly through Moonshot AI?

 

[jaden_yuki]

For now, the free tier is limited to 20 requests per day and 5 requests per minute. Could you tell me which exact Kimi version gives good results (is it Kimi-K2 0905 or Kimi-K2-Thinking)? do you use it via the OpenRouter API or directly through Moonshot AI?

You must be registered to see the links

 

[HikariSomnium]

Google has once again imposed total censorship on all models (it’s unclear why they even implemented “harmful” content filtering settings in the first place). It makes no difference whether it’s the free tier or the paid API - it’s a complete ban.


So it looks like only Kimi K2 and DeepSeek v3.2 remain. By the way, I tried Kimi K2 0905 - it translates well and feels somewhat more literary than DeepSeek v3.2, but Kimi K2 0905 doesn’t use chain-of-thought, so in cases where the context is too complex or there are other factors that require some reasoning for an accurate translation, the results aren’t very good.

It’s really a shame. I specifically prepared a bunch of small projects so I could translate them using two new models at once -GPT-5.2 and Gemini 3 Pro. I hope Google will reconsider their approach, but I definitely won’t deal with OpenAI anymore -they’re real thieves.

 

[jaden_yuki]

Google has once again imposed total censorship on all models (it’s unclear why they even implemented “harmful” content filtering settings in the first place). It makes no difference whether it’s the free tier or the paid API - it’s a complete ban.


So it looks like only Kimi K2 and DeepSeek v3.2 remain. By the way, I tried Kimi K2 0905 - it translates well and feels somewhat more literary than DeepSeek v3.2, but Kimi K2 0905 doesn’t use chain-of-thought, so in cases where the context is too complex or there are other factors that require some reasoning for an accurate translation, the results aren’t very good.

It’s really a shame. I specifically prepared a bunch of small projects so I could translate them using two new models at once -GPT-5.2 and Gemini 3 Pro. I hope Google will reconsider their approach, but I definitely won’t deal with OpenAI anymore -they’re real thieves.

how did DeepSeek v3.2 perform in your tests compared to Kimi K2 0905?

 

[HikariSomnium]

how did DeepSeek v3.2 perform in your tests compared to Kimi K2 0905?

Not very well. The wording feels kind of strange - it seems like it tries to be overly literal, and that doesn’t look great. However, there are places where Kimi K2 0905 kinda failed to grasp the core idea from the context, while DeepSeek v3.2 handled it correctly.

Spoiler: Here’s an example of a short snippet

You don't have permission to view the spoiler content. Log in or register now.

Overall, for my current small projects, I was already planning to use two translations from different models at the same time. The two models are meant to complement each other (with Kimi as the primary model and DeepSeek as a secondary one, used when something is unclear or when I have doubts). For relatively larger projects (starting from small visual novels with 5+ hours of gameplay), I would definitely use DeepSeek v3.2, since it allows you to save on input token caching.

Spoiler: UPD

You don't have permission to view the spoiler content. Log in or register now.

 

[jaden_yuki]

Not very well. The wording feels kind of strange - it seems like it tries to be overly literal, and that doesn’t look great. However, there are places where Kimi K2 0905 kinda failed to grasp the core idea from the context, while DeepSeek v3.2 handled it correctly.

Spoiler: Here’s an example of a short snippet

You don't have permission to view the spoiler content. Log in or register now.

Overall, for my current small projects, I was already planning to use two translations from different models at the same time. The two models are meant to complement each other (with Kimi as the primary model and DeepSeek as a secondary one, used when something is unclear or when I have doubts). For relatively larger projects (starting from small visual novels with 5+ hours of gameplay), I would definitely use DeepSeek v3.2, since it allows you to save on input token caching.

Spoiler: UPD

You don't have permission to view the spoiler content. Log in or register now.

since the lack of reasoning seems to be the issue with Kimi K2 0905, you should try Kimi-K2-Thinking

 

[HikariSomnium]

since the lack of reasoning seems to be the issue with Kimi K2 0905, you should try Kimi-K2-Thinking

Something isn’t working properly with Kimi K2 Thinking. This happens with some Thinking models. Processing a small 30-row CSV file took nearly 25 minutes. It burned through credits at a rate tens of times higher than regular Kimi, and the file still didn’t get translated ¯_(ツ)_/¯