.@ionet is at ~165B tokens through @OpenRouter in the first 13 days of October. That’s already ~1.8x everything we did in July. If the current pace holds, October lands somewhere around 4x July’s total volume. I care more about that than raw GPU count tbh. The useful output per deployed GPU keeps moving as sparse models get better and the serving stack gets tighter. So the network can grow in two directions at once: more hardware coming online, and more work getting squeezed out of what’s already there. Tokens served per GPU is a great sign of whether the network itself is actually getting better or not.
brandNo recent capture
OpenRouter on X
The conversation is led by 35 people describing their own OpenRouter setup working, with 31 quoting its token and spend figures as the industry's usage signal. The sharpest complaint is the fallout from the Railway breach, where keys leaked or were disabled.
First written 27 Oct · what changed · from 243 posts by 233 people
Tracking posts that say “openrouter”
Track this for your marketPosts, last 24h
—
Last captured 36h ago
Posts, last 7 days
388
123 in the last 72h
Velocity
55/day
7-day average
Momentum
—
Needs a recent capture to measure
Posts per day, last 30 days
as of Nov 3, 11:01 UTCWhat changed
as of 27 Oct · checked once a monthFirst written 27 Oct. When what people say changes, it shows here.
Tone· 135 posts from 132 people· written 27 Oct
How do people feel about OpenRouter?
Sentiment is positive: the largest group, 35 people, describe their own setup working on OpenRouter, and 31 treat its token and spend figures as the clearest read on what developers use. The 13 who are fed up point at keys leaked or disabled in the Railway breach.
The setups people name are ordinary and specific: GLM 5.3 or DeepSeek for cheap work, Opus 5.5 or Qwen 3.8-Max-Prime for harder coding, driven from Claude Code, OpenCode or a Hermes harness on one key. Some credit the company directly — one post thanks it for moving quickly to protect customers after the breach, another says "openrouter byok made it easy" to switch.
The numbers theme is people reading the platform from outside: posters cite Chinese models at roughly 57 trillion tokens a week against 16 trillion for US models, Anthropic's share of tokens against OpenAI climbing from about 20% to 50% after Opus 5.5, and open-weight models taking about 72% of tokens. 26 talk about cost from their own bills — $10 loaded in August with $9.23 left, free models turning out real work, the ":free" models used to trial a model before committing.
28 point others to it as the safe first stop, for cheap APIs, free models or abliterated ones. At 13, frustration is a minority: a missing cache discount, credit eaten mid-task, deployments broken by key rotation.
- Happy with it35
- Its numbers are the industry signal33
- Recommends it to others28
- Good value for the money26
- Frustrated with it13
See 5 of 135 postsHide the posts
I loaded $10 into openrouter on 1st Aug. Since then I've used Hermes as a personal assistant (RPi + Telegram stack) with DS Flash/Luna/Haiku class models. I've got $9.23 left today My usage is mostly shopping lists & lookup stuff but this is so cheap
Qwen 3.8-Max-Prime.. Openrouter.. Incredible.. If your coding anything. Give it a try.. I mean its like Fable x2.. I am shocked. When did this newest iteration get released? Incredible. It just fixed all of Fables code.. in minutes from the last 48 hours.. @Alibaba_Qwen we do not own enough Alibaba stock... period..
we have openrouter, gemini, codex, grok, idk if that's really "not proper"
Openrouter rotated the key. Railway can't build. 6 hours my app is down, i am shedding money ffs
5 of the 135 posts behind this answer, one per person. The other 130, and every new one as it’s posted, come with tracking. Get early access
Pain points· 73 posts from 69 people· written 27 Oct
What do people complain about with OpenRouter?
The largest problem, 14 people, is keys that leak or get abused: after the Railway breach, calls owners never made ran up charges on their keys. Next, 13 people question what OpenRouter's figures actually measure, and 9 had keys disabled by OpenRouter and could not redeploy.
Key abuse is told with numbers attached: Opus 5.5, Sonnet and Fable calls a key's owner never made, one loss stopped at about $7 once OpenRouter blocked it, half a month of credits taken. The lesson posters pass on is a spend limit per key — "the limit is the only reason this is a funny screenshot and not an expensive one".
The doubt about figures is about meaning rather than error: that token share is not revenue share, whether cache hits are counted, and that a usage leaderboard rank is not capability. 9 describe a catch-22 — OpenRouter disabled keys it thought had leaked, and Railway had paused deployments, so new credentials could not go live; one app was down six hours. One post credits OpenRouter for emailing promptly.
Smaller complaints: credit drained by a client defaulting to an expensive model or hitting zero mid-task; prompts passing a middleman, with NEAR AI saying requests through a router fall outside its confidential environment; free-tier caps of 20 requests a minute and 50 a day.
- Stolen keys and charges14
- Its numbers questioned14
- Forced key rotation9
- Errors, downtime and bad behaviour7
- Credits run out or drain fast6
- Router sees prompts6
See 5 of 73 postsHide the posts
railway just got compromised and this one might affect YOU directly if you use openai, claude, stripe, or openrouter and your API keys were stored on railway those keys may have been exposed attackers with your API keys can run up charges on your openai and claude accounts, process payments through your stripe, and route requests through your openrouter railway says they’re investigating but here’s what you should do right now if you used their platform: rotate every API key that was stored on railway. don’t wait for their full report. rotate now check your billing on openai, stripe, and any other service connected through railway for unexpected charges revoke old keys and generate new ones. the old keys should be treated as compromised “we can confirm all our customers’ money and data are safe” is the same line every company uses right before we find out they weren’t secure your keys first. read the report later
PSA: put a spend limit on every api key you create this is my openrouter usage the day my key got exposed. the limit is the only reason this is a funny screenshot and not an expensive one
Openrouter rotated the key. Railway can't build. 6 hours my app is down, i am shedding money ffs
72% of tokens on OpenRouter going to open-weight models is worth watching. token share isn't revenue share though
I've normally had okay experiences with Copilot but even that's turned to shit It randomly decided it wanted to default to an expensive model on OpenRouter and drained all my credits. I changed the provider to something I have more credits with and it went back to OpenRouter ://
5 of the 73 posts behind this answer, one per person. The other 68, and every new one as it’s posted, come with tracking. Get early access
Alternatives· 34 posts from 33 people· written 27 Oct
What are the alternatives to OpenRouter?
The largest comparison, 7 people, weighs calling the labs' own API keys directly against routing everything through OpenRouter. 5 use OpenRouter to keep working after Claude or Codex restricted their account, and running locally, vendor subscriptions and smaller inference providers are level at 4 people each.
Direct keys is a mix of habits and hedging: some run both ("usually direct or im through openrouter/vercel"), some want to be able to swap everything quickly, and one post points out that Anthropic models through OpenRouter still go back to Anthropic's platform and its safeguards.
The ban workaround is practical advice: use GLM 5.3 or Kimi K3 through OpenRouter rather than risk a Claude or Codex account, with one developer saying they can't use claude/codex any more "due to restricted cybersec capabilities". A poster notes OpenRouter picking non-China servers for DeepSeek.
The level themes: local, where posters price it out at 15x cheaper to use APIs than local inference, though someone running it monthly says a local machine still makes sense; vendor subscriptions, where a Kimi plan "isn't worth it" so the model is run through OpenRouter instead; and smaller providers, weighed on latency — 844ms on OpenRouter against 381ms elsewhere for one voice agent.
Two people compare LiteLLM as their own gateway with OpenRouter kept as fallback, and two discuss crypto inference providers' small share of volume.
- Direct API keys instead of the router7
- Bypassing lab account bans5
- Running models locally4
- Vendor subscriptions4
- Smaller inference providers4
- LiteLLM2
See 5 of 34 postsHide the posts
nothing changed through openrouter. are u using ds api directly?
Anyone using direct api keys from OpenAI, Anthropic, Google, DeepSeek, etc instead of OpenRouter? Getting worried about how fast I can be if I need to change everything in a short period of time.
🚨 To use REA without losing your Claude or Codex account Use GLM 5.3 or Kimi K3 through subscription or Openrouter, they're really good at reverse engineering Do NOT use REA on your Claude or Codex account! You will get banned in a couple of hours
I'm finally at the stage of dev where I can't use claude/codex anymore due to restricted cybersec capabilities. have to finally plug in to openrouter to build the thing that's going to save me from having to ever do it again
ok, asked gpt for a new calculation, assuming 10x input/output ratio and 90% cache reads. Openrouter prices. its 15x cheaper to use apis instead of local inference.
5 of the 34 posts behind this answer, one per person. The other 29, and every new one as it’s posted, come with tracking. Get early access
Questions· 12 posts from 11 people· written 27 Oct
What do people ask about OpenRouter?
The dominant question, from 6 people, is whether a newly launched model can be used through OpenRouter yet. The rest is scattered: two ask why an 18+ check appears before a coding task, two ask whether it is good enough, and two want a per-model cost breakdown.
Availability questions come in two forms: users tagging a launch to ask "when openrouter?" or whether it will be available, and posters asking whether a vendor's models are reachable through it at all. It reads as a wish to try something on the key they already have rather than set up a new account.
The age check is asked as surprise rather than complaint: one user had to confirm they were 18+ before asking Muse Wonder to redesign a website about toilet paper, and said it was a first across the models they had tested.
The cost questions are about what a bill is made of: whether OpenRouter breaks usage down per model, so a jump in tokens can be attributed to a cheap model running many agent steps under a planner, and how a model listed far below the stated inference price can be profitable. The two "is it good enough" posts are one asking whether the router is adequate and one asking how its pricing compares to another router.
- Will this model be on OpenRouter?6
- Why the age gate?2
- Is OpenRouter good enough?2
- A per-model cost breakdown2
See 5 of 12 postsHide the posts
Best model in for personal agents, yet we can’t use it in our personal agents today?!?! In all seriousness, congrats on the launch. Will at least one of two come to OpenRouter soon?
Will it be available on @OpenRouter ?
@grok Do openrouter and groq provide nvidia models, do nvidia provide the inference directly, wondering why they are not used extensively.
I asked Muse Wonder to redesign a website about toilet paper. First, I had to confirm I was 18+ via OpenRouter. I've tested plenty of AI models. This was a first. An age check before a coding task? I wasn't expecting that.
3.6x on a 7 day average is +260%. the split matters more: if most of it is haiku 5.5 running agent steps under an opus planner, that's a pricing story. does openrouter break it down per model?
5 of the 12 posts behind this answer, one per person. The other 7, and every new one as it’s posted, come with tracking. Get early access
Wishes· 7 posts· written 27 Oct
What do people want from OpenRouter?
Each of the 7 people here asks for something different, so no request has more than one backer. Among them: pricing per finished task rather than per token, IP allow-lists on every tier, and a router serving only uncensored models.
The cost ask is about agent runs: "price per token is half the story on long agent runs", since a model that loops twice costs more, so the poster wants cost per finished task shown somewhere. The access ask is a one-line request to make IP allow-listing available on all tiers.
Two asks are about trying a model on OpenRouter's terms: one wants an "openrouter but for uncensored models only", another wants a model dropped on OpenRouter so people can try it instead of gated behind a subscription. Smaller asks: support for x402 payments for API calls, crowdfunded inference capacity where a donation could be traced to prove the compute was used, and better default settings when serving image models — a request one post says has already been taken to OpenRouter.
- Cost per finished task1
- IP allow-lists on every tier1
- An uncensored-only router1
- Add the model we want1
- x402 payments1
- Crowdfunded compute1
See 5 of 7 postsHide the posts
price per token is half the story on long agent runs, a model that loops twice costs more. will you show cost per finished task anywhere?
@OpenRouter Make IP allow-list available to all tiers
@dealignai I need a openrouter but for uncensored models only
Drop it on openrouter for people to try it. Claude and OpenAI has never gate kept their models behind their subscription, why in the world would anybody trust that your product is better when it is very unproven, and has had usage limit concerns when it dropped. Like you guys came back in ai, but still make insanely dumb decisions like this. You can’t bring the meta subscription to other harnesses like opencode either, so I’m just having a hard time rationalizing spending money on a sub to use one model that is not better than opus 5.5.
Openrouter for x402 could really boost API efficiency now
5 of the 7 posts behind this answer, one per person. The other 2, and every new one as it’s posted, come with tracking. Get early access
In their words
How do people talk about OpenRouter?
The phrases people use most, counted once per person.
- 1“through openrouter”9
- 2“openrouter key”5
- 3“open-weight models”3
- 4“providers on openrouter”3
- 5“spend limit”2
- 6“see your prompts”2
- 7“instead of openrouter”2
- 8“free tier”2
- 9“per token”2
- 10“api directly”2
- 11“agent loops”2
- 12“uncensored models”2
Posts are shown the way X shows them. One of them yours? Here’s why it’s shown, and how to take it off.
Also in Inference and model APIs
- Baseten−25%Approval of how fast Baseten serves open models and how quickly people get running leads the talk, with a free window on its mammoth-1 and elephant-1 decision models drawing recent posts.366 posts / 7d
- Modal+31%The largest group on Modal, 46 of 96 people, posts positive reports, many about setup disappearing rather than one feature.325 posts / 7d
- Groq−56%Builders credit Groq for what they shipped on it — the largest group, 21 people — with the free tier at 19.320 posts / 7d
- Fireworks−60%The talk is about speed: a DeepSeek V4.1 Flash release on Ultrafast at 1,521 tok/s, with 13 people won over and 8 asking what tokens per second actually buy them.264 posts / 7d
Get every post about OpenRouter, filtered and labelled
This page is the summary. Tracking gives you the stream behind it for any brand, keyword or @mention: each new pain point, request and question as it’s posted, with the spam and promotions already taken out.