@petergostev Would you recommend it even for coding on a 20$ plus plan over Sol?
OriginalJ'ai testé un peu le modèle GPT-6.1-Terra. C'est probablement le premier modèle non-frontière qui franchit le seuil de quelque chose que ça ne me dérange pas vraiment d'utiliser. On dirait une version un peu plus simple de 6.1-Sol, ce qui est honnêtement plutôt bien. Ça n'a rien à voir avec ce qu'était le GPT-5.6-Terra, qui semblait un peu horssujet. Si vous êtes à court de quota, si vous avez besoin d'une tâche informatique de longue durée, ou si vous avez juste besoin d'un bon cheval de labour fiable, ça fera l'affaire. Je ne vais pas vous dire que ça va changer votre vie, mais
0
0
0
0
Commentaires 28
-
-
@petergostev Is it any faster tok/s wise? Thinking of getting back to a plus plan to try it out
Original -
@petergostev Thoughts about it as a daily driver for openclaw agents, unless its coding work etc??
Original -
@petergostev gpt 6.1 sol very good for 20$ budget and also good usage tbh
Original -
@petergostev felt like a simpler sol after a bit. what did it get wrong?
Original -
@petergostev And it is really fast as well, just tried with video editing
Original -
@petergostev 我只是不希望它像 sol 那样慢，太慢了太慢了，做不了任何工作
Original -
@petergostev “non-frontier model that crosses the threshold” seems very interesting, I think there will have lots of open source models with exact descriptions coming out and competing on cost efficiency.
Original -
@petergostev The best model isn't always the smartest one, it's the one you can afford to keep working. Terra sounds like a strong default for everyday tasks, while Sol still earns its place when the extra reasoning matters. This comment was written with Terra 6.1, using @CryptoReplyGuy1.
Original -
@petergostev Computer use? What for though?
Original -
@petergostev Aún estoy apenas empezando a probar 6.1 Terra y por ahora en 2 pruebas que hice contra 6.1 Sol ha sido de media 6 veces más rápido Terra!!!!! Que locura de rapidez⚡️
Original -
@petergostev Curious what you'd still hand to Sol instead
Original -
@petergostev So basically tell 6.1 sol to plan and verify 6.1 terra as his worker?
Original -
@petergostev For 20$ is it going to replace Deepseek-v4.1-flash for me? better quality?
Original -
@petergostev That’s a meaningful shift if it feels good enough for everyday use without constantly needing the frontier model oneself
Original -
@petergostev agreed on the sol vs terra take. in our code gen test terra built the whole scene for $0.95 in 31 minutes. opus was prettier light but 5x the cost. for most workloads sol is overkill now https://x.com/thehypedotnews/status/2114799954619490582?s=20
Original -
@petergostev 6.1 terra feels like it was optimized to run dots for $20 / month users
Original -
This might be Terra's biggest win: becoming the default workhorse rather than another benchmark champion. For long-running agents, stretching usage limits matters enormously. Have you noticed any difference in how Terra handles blockers, retries, or context-heavy tasks compared to Sol?
Original -
@petergostev I've been using GPT-6.1-Terra for a bit, it did a lot of work but the 5hours limit drained fast as usual. but in the other hand it accomplished a lot of what i asked for or partially smh
Original -
@petergostev Agreed Peter . . . Funny how parts of X judge a model by how well it plays Minecraft, while dismissing what matters to businesses: reliability, cost, and getting real work done. If you care about applied AI, Terra is a big deal. Not every breakthrough needs to win a video game.
Original -
@petergostev it's promising in terms of pricing, but it's very lazy https://x.com/atomic_chat_hq/status/2114811032816476638?s=20
Original -
@petergostev Solid workhorse
Original -
@petergostev Im having a hell of a time trying to use terra 6.1 it just keeps giving up on its own tasks and refusing to complete them.
Original -
@petergostev Opus 5.5
Original -
@petergostev "Don't mind using" is an underrated bar. Most people don't need the best model, they need one that doesn't make them check every answer twice.
Original -
@petergostev 속도가 빠르고+작은단위로 시킨것만 잘 한다면 최고의 모델 아닐까요
Original -
@petergostev 模型选择越来越务实了。日常任务用便宜、稳定的模型，难题再交给最强的模型，长期算下来可能更划算。以后拼的不是谁只用一个顶级模型，而是谁更会分配任务。
Original -
@petergostev Doesn't seem like you're qualified to give that advice tbh. Did you not see the terminal bench graph?
Original
스크롤하면 더 불러옵니다