Codex 28 · Theme · #17
Codex token throughput
27people describe this
1 on Rantr · 26 on X
X replies to the promise post are counted apart from Rantr votes. “This happened to me too” adds your browser to this theme on Rantr, once.
Where OpenAI stands
Not addressed yetNo recorded announcement has answered this yet.
In their words
What people said
- Problem
Honestly, I used to pray for ChatGPT to beat Claude, but now I just need Gemini to blow GPT out of the water. Codex capping at ~40 tok/s while burning through limits in a blink is literally wasting our lives—I can't ship real projects like that. Meanwhile, Gemini 3.8 Flash is…
@dikombehero on X ↗ · 23 Oct 2026
- Problem
So what happened to the 28 days? Anything useful coming, or just junk? Still stuck at 25 tokens/sec.
@OerxRbat on X ↗ · 19 Oct 2026
- Problem
Dude 6.1 sol’s speed is back to where it was before your day 1 speed bump. It’s really untrustworthy to back track upgrades silently.
@lumi_inv on X ↗ · 15 Oct 2026
- Problem
Why is the GPT 6.1 sol model I'm using on Codex so slow, dropping to just a few tokens per second? Didn't you say it was updated to 50 tokens per second? If it's this unstable, what kind of an update even is this?😅
@sleeplyzoe on X ↗ · 12 Oct 2026
- Problem
根据目前的测试，大约在 35 TPS，无法达到 50 TPS，我是 200 美元 pro 用户
@zhudekui5656 on X ↗ · 11 Oct 2026
- Problem
Average speed on Sol-6.1 is still around 20 TPS. Kinda dissapointed ngl.
@DawidRKnaz on X ↗ · 10 Oct 2026
- Problem
gpt-6.1 sol in Codex is still painfully slow, barely 20+ TPS. How does adding dots to mobile meaningfully help Codex/Work users? If there's no real improvement today, can we at least get the full reset as promised?
@LowWeikee on X ↗ · 9 Oct 2026
- Mixed
50 TPS? Maybe only on your account, connected directly to the data center. 22-26 TPS
@diego_dion58017 on X ↗ · 9 Oct 2026
- Mixed
Currently running at 20TPS....
@illogicparadox on X ↗ · 9 Oct 2026
- Mixed
15 tokens per second rn for me
@James56009782 on X ↗ · 9 Oct 2026
- Win
They've fixed it, I was testing it on their Playground API (web version), now instead of 49 tok/s I'm getting 249 tok/s there. :)
@SimonasLTU1 on X ↗ · 9 Oct 2026
- Mixed
I’m getting ~150 tok/s in pi, what harness are you using?
@Chris_Falck on X ↗ · 9 Oct 2026
- Problem
Getting an average of 23 tokens per second after your "speed boost" from day 1 of "improvements"
@SharkThePark on X ↗ · 9 Oct 2026
- Problem
that is his definition of getting more work done. it's running for longer and drawing tokens slower. but in reality you're just waiting for shit. claude draws faster but actually gets your shit done. i have the same issue with codex right now. ive cancelled my 200 plan
@u83205 on X ↗ · 9 Oct 2026
- Mixed
因为标准速度已经被降到了 6 tok/s，8x 差不多就是 49 tok/s 了
@Plasticine9750 on X ↗ · 9 Oct 2026
- Problem
In the end, does anyone get 50 TPS? I only get 22.5 on average with Astra....
@xryul3 on X ↗ · 8 Oct 2026
- Problem
token speed improvement only works on day one? 6.1 sol in codex became 20 tps again, it maybe the worst speed across model providers.
@hsudatalks on X ↗ · 8 Oct 2026
- Mixed
Feels like Codex is doing 10 tokens per second right now. It's so good daam slow on my 64GB Macbook Pro
@iuditg on X ↗ · 7 Oct 2026
- Problem
I am still on 23 tps, this is excruciating.
@SKK1869512 on X ↗ · 7 Oct 2026
- Problem
I am still on 23 tps, this is excruciating.
@SKK1869512 on X ↗ · 7 Oct 2026
- Problem
@thsottiaux Please take a good look at its Tokens output efficiency, which is now ridiculously slow. This is a continuous task of more than 30 hours, and the Tokens output efficiency per minute is only less than 10.
@Xiaolizi_X on X ↗ · 7 Oct 2026
- Problem
I dont thing 6.1 get and speed !
@EmadAlslahi on X ↗ · 6 Oct 2026
- Problem
Astra is and always was fast for me. Sol 6.1 is still at 20 t/s or something, taking (over) 3 times as long for a task as Sol 6.0.
@MakJoris on X ↗ · 6 Oct 2026
- Mixed
I see 35 tps :) right now in openrouter is that an improved already
@ThnhNgu97660433 on X ↗ · 6 Oct 2026
- Problem
My speeds dropped from 28 tps to 10tps
@swg0101 on X ↗ · 6 Oct 2026
- Problem
Where the fuck is the promised DEFAULT 50 TPS? Three GPT-6.1 Sol runs hours later: 31.9, 32.1, 32.6 TPS. Same 440-token text, no tools, first-token wait excluded. Fast isn't default. Stop the bullshit victory laps. Show your benchmarks. Do you think we're idiots?
@V3zxS8myOU4056 on X ↗ · 6 Oct 2026
- Idea
From "your X community" 1. Remove the 5-hour usage limit on Plus for Codex and Work. 2. Make the Sol budget last a real session, not a single routine task. A 5-hour window should not die in half an hour of normal repo work. 3. Fix GPT-6.1 Sol speed. ~20 t/s on the standard plan…
@Rolandofj on X ↗ · 5 Oct 2026
Your experience
Had this too? Say what happened.
JEV reads your words as you type and shows which theme they belong to. It may place them in another theme if that fits better.
Status
Where OpenAI stands
“Shipped” means an announcement was linked to the theme by AI, not that the problem is fixed for everyone. “Not addressed yet” means no recorded announcement was linked; it is not proof that nothing changed.
No recorded announcement has been linked to this theme yet.
Corrections
Changes to this theme
-
“Silent speed rollback after upgrade” was merged into this theme: JEV judged them the same problem.
-
Renamed from “Low token output rate” to “Codex token throughput” as more people described it.
-
“Current token rate visible” was merged into this theme: JEV judged them the same problem.
-
“Slow token output over time” was merged into this theme: JEV judged them the same problem.
-
“Response speed varies by version” was merged into this theme: JEV judged them the same problem.
-
“Token rate stays low” was merged into this theme: JEV judged them the same problem.
-
“Initial token speed drops later” was merged into this theme: JEV judged them the same problem.
Share
Share this theme
Follow
Follow this theme by email
We send a confirmation link first; nothing else is sent until you confirm. Then at most one email a day, only when the theme changes.
X thread: 2769 replies to the promise post collected (4 Oct 2026 – 26 Oct 2026): 2555 placed in a theme and 214 set aside as off-topic or unsafe. X replies are found by search and each one is checked against its public post on X. Last read: 26 Oct 2026, 23:46 UTC. Not every reply is read.
← Every theme