I would say go with codex but the limits are also fluctuating. the past two weeks its been crap but thankfully they kept hitting a reset often enough. today has been amazing. I think I would be down to 50% of my usage but its doing pretty well and i am down like 8% while still using fast mode.
find ~/.claude/projects -type f -mtime -3 -regextype posix-extended -regex '.*/[0-9a-f]{8}(-[0-9a-f]{4}){3}-[0-9a-f]{12}\.jsonl' -printf '%f\n' | sed 's/\.jsonl$//' | sort -u
01a63270-08e2-4a76-9898-55016f6cc667
17003b79-5eb1-4849-be48-f8224bf90ca7
27cd5d41-41e3-4ded-b829-4fb4f3dda9cd
2af962ba-33e3-43f9-9560-3c6cc693e0b7
4c8f34a6-5447-4ab6-9873-e126f3ac5017
4fb0a8f2-3c32-47f9-9ebe-e3d2977ca745
5979c312-51b5-4a1f-b9d6-c7a136b97319
72a0deb9-a2cc-4624-827f-6e67dfd9e6b6
7ba10765-8388-4585-9505-53807f45b693
8520d668-2a56-4e85-a92e-2219cd727edf
ac47f1a3-c012-4307-af81-12b98d26563a
ba21e8a1-48a9-4567-b2f8-937bb8e54573
bcf4f652-7024-4ae3-b86f-8142f4a0df8d
bf89d814-6438-4fd3-9cca-9b7815a80b6c
c2f9ea37-5c6e-426e-9bc7-e4f5ff28b77f
c4fa3fb7-a95b-452e-9540-21685bfd01d8
cf96832f-0a79-4082-855c-d71b471f6500
d4954574-5da4-4dda-8b6f-0070ce50677b
d51ee6c7-d84a-41de-9905-8211f17bdfb2
d5593d13-f330-48a5-aa1a-65ff32336364
d9dde79c-392e-4094-ba61-007d5a924315
ead2756a-9b3b-4c72-b162-8af57dacccc9
fd7b1328-e243-4d57-812c-b2d9fea11c4c
this is a new account i just created and ate 50% of my weekly....
GPT 6.1 Sol is now on BridgeBench.
OpenAI is calling it "near-Astra intelligence."
Our results: 3 points better than GPT 6 Sol.
82 points behind GPT 6 Astra.
When a model jumps on the lab's own benchmarks but barely moves on independent ones, that's usually a sign of benchmaxing.