@refresh_dev

Computer Use & Coding RL Gyms

refresh.dev/
Joined August 2025
  • Tweets9
  • Following3
  • Followers407
3Media
Refresh retweeted
TechCrunch
@TechCrunchOct 5
Nvidia-backed Reflection AI unveils Beam, its first open-weight model, which it says rivals GLM-5.2 on reasoning with far less inference compute. Weights are due this month. spr.ly/6019BGLSCz

Reflection debuts Beam, an open-weight AI model to rival Chinese models at lower compute cost |...

Reflection is aiming Beam and future models at enterprises and sovereign nations. The pitch is to build “AI factories,” a product that would let institutions build their own customized, local AI...

techcrunch.com
34
24
5
161
39,446
Congrats to the @reflection_ai on the great model release. Exciting times for open intelligence!
Reflection
@reflection_aiOct 5
Introducing Beam: a highly efficient agentic open model with 501B total parameters and 23B active. - Frontier reasoning efficiency - Advances the Western open frontier on coding & agentic tasks - Trained end-to-end from scratch Full weights release this month. Learn more about Beam: reflection.ai/beam
1
9
680
Built 100% in @harborframework btw cc @alexgshaw, @ryan_marten , thanks for continuing to support such a great platform!
6
1,093
Also feel welcome to get in touch at [email protected] :)
1
4
1,162
Check out the full blog post here: refresh.dev/blenderbench DMs open if interested in collaborating on 3D LLM evals!

BlenderBench

Going beyond modeling with frontier models.

refresh.dev
2
10
1,408
Astra's 20 runs cost an estimated $55.23 vs $20.54 for Opus 5.5. Both sit on the cost-performance frontier. More spend isn't always better. On this task, Astra spent $2.27 and scored 0.200 after changing the protected ring; Opus spent $3.91 and scored 0.600.
1
7
1,498
GPT-6 Astra averaged 0.686 reward across 20 Blender tasks, Claude Opus 5.5 0.642. Astra hit 0.90+ on 7/20, Opus on 3/20. Where they fall short: on this tightrope task, Opus 5.5 kept the cowboy's feet planted instead of stepping and scored 0.400.
1
1
15
2,153
Introducing Blender Bench v1 (BBv1): a benchmark for frontier LLMs on 20 realistic 3D production workflows including animation, modeling, rigging, cloth & surfacing. GPT-6 Astra leads overall; Claude Opus 5.5 wins modeling & cloth simulation.
48
59
15
658
43,232
Refresh retweeted
Claude
@claudeaiSep 28
Introducing Claude Sonnet 5.5, the second model in the Claude 5.5 family. It’s a clear upgrade over Sonnet 5, runs more than 30% faster, and costs up to 30% less for most work.
1,966
4,464
3,032
55,843
12,761,289
Load more