SkalskiP
@skalskip92
Open-source Lead @roboflow. VLMs. GPU poor. Dog person. Coffee addict. Dyslexic. | GH: | HF:
1.4K Following 50.1K Followers
Opus 5.5 is the best "vision" model Anthropic ever made
better than Fable 5, better than GPT-6 Sol, worse than GPT-6 Astra
vs Fable 5.1 (high):
- estimated cost: ↓ 60%
- detection: ↑ 11.84 pp
- reasoning: ↑ 12.80 pp
vs GPT-6 Sol (high):
- extraction: ↑ 11.00 pp
- reasoning: ↑ 7.95 pp
↓ 16 more examples
Show more
Opus 5.5, GPT-6 Sol, and GPT-6 Luna numbers are on the Playground leaderboard
crazy how fast Google lost the computer vision lead
object detection used to be their stronghold, now 3 labs and 4 models sit ahead of them
link:
Show more
fuck me. every lab is working on vision. I have preliminary Opus 5.5 results. by far the best vision model from Anthropic. more soon.
I was so impressed 3 years ago…
RIP image annotation companies
Fully automated image labeling with GroundingDINO + SAM + OpenAI Vision API
code:
let's see how good it is at vision. running benches now.
Grok 4.7 is a strong combination of intelligence, speed & low cost
GPT-6 Astra is the best vision model we have tested
I put together our results on detection, segmentation, box prompting, counting, reasoning and video
full post:
↓ examples and trade-offs
Show more
I see so many “jev demos” claiming it “sees” something
correct me if I’m wrong. none of this is real
for now, jev cannot take image input
GPT-6 Astra is the best vision model we have tested
I put together our results on detection, segmentation, box prompting, counting, reasoning and video
full post:
↓ examples and trade-offs
Show more
SAM3 vs Astra for text-prompt segmentation
SAM3 gives much more precise masks. it just often fails to understand complicated text prompts.
Astra is great at language. it rarely misses what you want to detect, usually only when the task needs internal domain knowledge. masks come back as JSON, point by point. slower, and less accurate than SAM3
we just rolled out Astra segmentation in playground
link:
↓ more Astra examples below
Show more
Astra can do segmentation
this is pure VLM result. no expert models (like SAM) were used
No other VLM even comes close to this quality
- high effort
- avg input tokens / image: 2,052
- avg output tokens / image: 4,685
- avg cost / image: $0.255
- median time / image: 78.3 s
↓ GPT-6 Astra segmentation deep dive
Show more
going live in 20 min on Astra for vision
5pm CEST / 11am ET / 8am PT
detection, segmentation, visual prompting, auto-labeling, benchmarks, pricing, latency
join the stream:
Show more
basketball AI (95% local AI + 5% GPT-6 Astra)
- detect ball and players
- track players
- re-identify players across plays
- OCR player numbers
- recognize player in possession
- detect court keypoints
- map player positions and trajectories
Astra is powerful, but expensive tool. use it when it actually makes a difference.
↓ deep dive
Show more
is Astra really this good at VLA?
I spent hours trying a zero-shot demo with the SO-101 arm and this is all I got
i gave astra a robot, a paint brush, and a camera then asked it to paint the golden gate bridge in real life!
it figured out how to control the robot, and progressively got better throughout its attempts. the timelapse is sick
Show more
“no one is forcing anyone to code with ai”
if you want to be competitive on job market you need to embrace agentic coding. otherwise you are 10x less productive. just a reality.
I loved writing code by hand. I miss it.
I loved writing code by hand. I miss it.
supervision crossed 50,000 GitHub stars
5,131 commits, 1,732 PRs, 188 contributors
thank you for building this with me
link:
prompt GPT-6 Astra with boxes to generate new detections
- green - positive examples
- red - negative examples
- blue - new detections
prompt GPT-6 Astra with boxes to generate new detections
really useful when you know the object, but can’t describe it well in words (like shipping container wall dent)
best part is you can do this between different images
- green - positive examples
- red - negative examples
- blue - new detections
↓ GPT-6 Astra box prompting
Show more
basketball AI (95% local AI + 5% GPT-6 Astra)
- detect ball and players
- track players
- re-identify players across plays
- OCR player numbers
- recognize player in possession
- detect court keypoints
- map player positions and trajectories
Astra is powerful, but expensive tool. use it when it actually makes a difference.
↓ deep dive
Show more
basketball AI with RF-DETR, BoT-SORT and GPT-6 Astra
- RF-DETR: local player and ball detection
- BoT-SORT: local player tracking
- GPT-6 Astra: player ReID, player recognition, ball possession
cost: $0.2 / 1s video
latency: 16s / 1s video
more soon
Show more
We want to celebrate with people using GPT-6. We did this for GPT-5.5 and it was really fun.
We’re getting together in SF on September 16 to talk about the model, what we should build next, and mostly just to hang out.
Apply by Sep 10:
Show more
prompt GPT-6 Astra with boxes to generate new detections
really useful when you know the object, but can’t describe it well in words (like shipping container wall dent)
best part is you can do this between different images
- green - positive examples
- red - negative examples
- blue - new detections
↓ GPT-6 Astra box prompting
Show more
Astra really feels like “ChatGPT moment” for computer vision
updated vision model tier list with GPT-6 Astra and Gemini 3.8 Flash
Gemini dethroned
GPT-6 Astra is the best "vision" model I've seen
- notices tiny details
- draws conclusions
- produces precise and tight boxes
- sees and reads text in different contexts
- knows a lot about a lot
- fast
- expensive
↓ GPT-6 Astra detection deep dive
Show more
