I *really* hope a new CodeGemma with FIM support is one of the things in the oven! 😀
GENERATIVE AI
-
Flowise: Building LLM Applications Visually with No-Code
By
–
1/ Beginnings Flowise began with a simple vision – to help people build LLM applications visually using frameworks like @langchain and @llama_index
. We launched publicly in April 2023, and it immediately resonated with the developer community. -

Generating one million instances cost implications and impact
By
–
bon si t'en genère 1 million c'est pas cher mais ca va faire mal
-
GPT-4.1 outperforms GPT-5 on code completion speed
By
–
One simple thing to try – use GPT-4.1 as a basic code completion model; you'll see it's really fast and accurate. Try to get any size GPT-5 with any amount of reasoning to match it. In my testing, GPT-5 was ~5x slower at the point it matched performance.
-
GPT-4.1 and Sonnet 4 Adoption for Development Work
By
–
Thanks Ted! Tbh we've got back to using GPT-4.1 for our coding quick fixes, and Sonnet 4 / Opus 4.1 for our main work. So we won't have too much feedback until GPT-5 has improved enough for us to give it another try. Definitely barracking for you guys to succeed!
-

Google releases Gemma 3 with 270 million parameters
By
–
270 million (not billion) parameters! https://
huggingface.co/google/gemma-3
-270m
… -
GPT-5 GB200 FP4 Precision and Token Speed Analysis
By
–
So GPT-5 is running on GB200, to take full advantage of it, it could be in FP4. So I wonder if this change is hiding the size of the model somewhat. If it was FP8 and on H100, it would be now running at something like 10 tokens per second rather than current 50 tps.
-

Google Releases Gemma3 270M Tiny LLM on KerasHub
By
–
Google just dropped a new tiny LLM with outstanding performance — Gemma3 270M. Now available on KerasHub. Try the new presets `gemma3_270m` and `gemma3_instruct_270m`!
-
OpenAI’s GPT-5-Pro: Cost Optimization Over Capability Push
By
–
Given that we saw OpenAI test better checkpoints on the arena, it does also seem like they have also optimised for cost / mass appeal rather than pushed the capabilities as far as they could. We also don't have almost any benchmarks for GPT-5-Pro, which is genuinely superb and