The unified interface for LLMs. 400+ models (incl. 50+ free), the best prices, explorable data, private chat, & a unified API. openrouter.ai/discord
- Tweets 5,629
- Following 385
- Followers 150,175
- Likes 5,929
Jev is quickly beocming the primary choice for classification requests on OpenRouter.
It took 27% of weekly request volume in this category, nearly 2x the share of DeepSeek V4 Flash which previously held the top spot
32
40
458
35,518
You can see Jev Router's thinking process in OpenRouter Chat.
The routing insights panel shows the model it picked for each turn and the scores behind it, like task, difficulty, precision, and larger model benefit.
Test Jev Router here: openrouter.ai/chat?models=ty…
3
1
42
15,390
On five agent benchmarks, Jev Router had a faster median time to first token than every other router we tested.
1
1
66
22,154
On four agent benchmarks, Jev Router solved 82% more tasks than our Auto Router (237 vs 130 of 423).
5
2
115
31,037
Introducing typesafe/jev-router: a cache-aware model router powered by Jev and @typesafeai
The Jev Router picks the best model and reasoning effort for each request, balancing quality, speed, and cost.
Here's how it works 👇🏻
103
150
2,029
1,284,050
Final standings:
1. Qwen Image 3 Pro, 35%
2. Nano Banana Pro, 33%
3. Riverflow V2 Pro, 23%
4. GPT Image 2, 9%
The grand final was A Nano Banana Pro, B GPT Image 2, C Qwen Image 3 Pro, D Riverflow V2 Pro.
2
1
15
6,188
The people have spoken.
Qwen Image 3 Pro wins the first Memebench bracket, taking the “Why would I deceive you?” meme with 35% of the vote. Nano Banana Pro finished one vote behind at 33%.
Congratulations Qwen Image 3 Pro!
14
3
96
13,782
Once a month --> once a week --> every single day.
Keeping up with new model launches is a full time job.
26
32
521
26,554
Presenting the Memebench Grand Finals
The top 4 models from round 1 face off to determine which model cooked harder on the "Why would I deceive you" meme.
Which model cooked harder? A, B, C or D?
👇🏻 Vote in the poll below:
4
2
24
10,620
Web Search:
By default, models use native search on frontier models, or the best third party search providers for open weight models.
You can pin Exa, Parallel, or Perplexity ($0.001 to $0.007 per search), or bring your own Firecrawl key.
openrouter.ai/tool/web-searc…
1
3
26
7,353
Introducing the Server Tools Marketplace
A one-stop shop for better ways to ground your agents, including web search APIs, Shell, tool search, and more 👇
18
19
296
62,647
Meet Memebench - Group B
We gave 8 AI models the same meme and asked them to put @pingToven and @Audrey_Sage_ in it
Second round, group B.
Which model cooked harder? A, B, C or D?
👇🏻 Vote in the poll below:
9
1
27
16,848
Meet Memebench
We gave 8 AI models the same meme and asked them to put @pingToven and @Audrey_Sage_ in it.
First round, group A. Group B comes tomorrow.
Which model cooked harder? A, B, C or D?
👇🏻 Vote in the poll below:
15
4
52
16,457
Tip: learn how to improve your AI model observability using Logs, Activity, and Broadcast.
Includes a full matrix of all fields supported, per destination, in Broadcast, and APIs for fetching data.
openrouter.ai/docs/guides/fe…
9
57
9,558
GPT-6 Sol and GPT-6 Luna from @OpenAI are live on OpenRouter!
Half the price of their GPT-5.6 predecessors, Sol at $2/M input and $10/M output, Luna at $0.10/M input and $0.50/M output. On AutomationBench each one tops its predecessor's best score at a fraction of the cost per task.
What each one is for 🧵
17
18
248
16,082
Claude Opus 5.5 from @AnthropicAI is live on OpenRouter!
The first model in the Claude 5.5 family leads Opus 5 and Fable 5.1 on agentic coding, knowledge work, and computer use, with 1M context at $4/M input and $20/M output, 20% lower per token than Opus 5.
ALT Benchmark table comparing Claude Opus 5.5 with Fable 5.1, Opus 5, GPT-6 Astra, and GPT-5.6 Sol on agentic coding, knowledge work, business workflows, multidisciplinary reasoning, scientific research, computer use, and chart recognition. Opus 5.5 leads on seven of nine rows.
20
19
288
16,457
3/ Bigger batches don't mean much longer waits. One request finished in 5 to 11 minutes, 1,000+ requests in 12 to 21, depending on the hour you submitted.
1
7
1,500
2/ Most batches come back in minutes, not 24 hours.
Across 230k+ batches completed during the two week beta, the median batch finished in 7 minutes and 90% finished within an hour.
Submit outside 5 AM to noon Pacific and even the slowest tenth finish in about one hour.
1
21
2,538
1/ The OpenRouter Batch API is live.
For work that doesn't need an answer right now, the Batch API cuts per-token pricing in half on most models and returns everything within 24 hours.
Get started: openrouter.ai/blog/announcem…
22
21
353
56,280
TIP: automatically classify your LLM requests using @typesafeai's Jev, using Classifiers: openrouter.ai/workspaces/def…
You can then analyze the results in Explore: openrouter.ai/activity/explo…
12
6
158
13,951
Q: Which models have been most impacted by the arrival of Jev?
A: The flash versions of models across many labs.
Also: nearly half of Jev users on OpenRouter hadn’t touched any models the week before. This launch sparked enough excitement to pull them off the sidelines.
18
6
212
16,130
Xiaomi MiMo-V2.6 from @XiaomiMiMo is live on OpenRouter.
Three new models. A 1T+ parameter flagship, an open-source MoE, and a ~10x faster variant of the flagship. All three take text, image, video, and audio with 1M context.
More in the thread 🧵
42
53
1,059
71,168
6/ 🏆 jev_search
Natural-language search over a codebase in two passes. Jev first picks the likely files and shows those, then sweeps the rest.
Jev's scores: integration 2.4/4, interest 2.2/4
🔗 github.com/caio0452/jev_sear…
3
1
36
10,192
5/ 🏆 Vibe Domain
LLMs brainstorm domains that are already taken. Vibe Domain scans thousands of actually available domains and lets Jev judge which fit the vibe, required words and TLDs you want.
Jev's scores: usefulness 2.9/4, integration 2.8/4
🔗 obstudio.org/tools/vibe-doma…
2
1
57
11,575
4/ 🏆 tisco
A natural-language footage organizer. Transcribe 50 videos, hand the transcripts to Jev as state, then ask "which have relevant dialogue?" or "which are only rumbling and silence?"
Jev's scores: integration 3.3/4, interest 2.7/4
🔗 github.com/cairodavila/tisco
1
1
47
12,297
3/ 🏆 Jev Chess
One board, and the whole internet plays one game against Jev.
Every legal move is one Choice question, so illegal moves are impossible. Pieces shaded thru Probabilities in real-time
Jev's scores: integration 3.4/4, innovation 3.0/4
🔗 jevchess.com
3
3
56
13,901
2/ 🏆 JevAI for XMage
Magic: The Gathering opponents where every decision is a question to Jev. Code lists legal plays, Jev picks. A hybrid lets XMage search and Jev break ties. 11-6-3 vs the stock bot.
Jev's scores: evidence 3.4/4, quality 2.7/4
🔗 github.com/ShiftSad/mage/tre…
1
50
15,462
The addressable market for Decision models like Jev might be very large
Chart from openrouter.ai/rankings#task-…
20
23
309
22,994
Grok 4.7 from @SpaceXAI is live on OpenRouter!
Their most capable model for coding and knowledge work. It works longer on hard tasks, checks its own work more carefully, and sits at the frontier of price-performance on CursorBench 4.0.
Use it now: openrouter.ai/x-ai/grok-4.7
13
14
313
16,151
API key limit notifications live on the key itself.
Set a credit limit on a key, turn on spend-limit alerts, and get an email at 80% and 100% of that limit (or your own thresholds).
Alerts go to the key owner by default, or send them to workspace admins, org admins, or others.
1
5
2,910
Workspace budget alerts tell you when a workspace crosses a share of its budget.
Defaults are 80% and 100%, and you can add your own thresholds. Send to org admins, workspace admins, specific people, or a webhook.
1
5
561
Get the latest model updates from OpenRouter when they drop.
- Deprecation: a model you recently used is scheduled for removal
- Price drop: a model you recently used got cheaper
- New model: a new model just landed on OpenRouter
1
13
1,369
Stay on top of model changes, low credits, workspace budgets, and API key limits with Notifications in OpenRouter!
Pick which events notify you and how they're delivered 🧵 openrouter.ai/settings/notif…
11
3
85
20,339
3/ Jev came in second cheapest, just behind Qwen3.8 Flash, and under the three other LLMs.
Typesafe's thesis is that decision models are fast, accurate, and most importantly, cheap.
On this task, the cost came in cheaper than 3 of 4 models.
ALT Bar chart of cost per 1,000 decisions measured from OpenRouter usage. Qwen3.8 Flash $0.040, Jev $0.049, DeepSeek V4.1 Flash $0.075, GLM 5.3 Flash $0.083, GPT-5.6 Luna $0.179.
2
2
52
4,620
2/ A question we saw many ask: does accuracy suffer for the speed?
On this task, no.
Jev matched the most popular LLMs used for classification, and all five landed within a handful of cases of each other. A decision model was as correct as a full LLM.
ALT Bar chart of accuracy on 200 task-classification cases. Jev 98.5%, DeepSeek V4.1 Flash 98.5%, Qwen3.8 Flash 97.5%, GLM 5.3 Flash 97.0%, GPT-5.6 Luna 96.5%.
7
2
71
5,339
1/ Jev, a decision model by @typesafeai, sparked a burst of projects and discussion. We tested it using Ori Eval against popular LLMs on OpenRouter at judging.
Jev was >5x faster than the next fastest model, and even its slowest requests beat every other model's median.
ALT Bar chart of median latency per model. Jev 154 ms, GPT-5.6 Luna 860 ms, DeepSeek V4.1 Flash 911 ms, Qwen3.8 Flash 914 ms, GLM 5.3 Flash 1544 ms.
77
62
969
60,065
What is a decision model?
Jev by @typesafeai answers yes/no and multiple-choice questions, with a confidence score. Much of software development are a sequence of decisions, and Jev is 10x cheaper and faster than an LLM.
Let’s understand this through practical examples:
68
125
1,692
168,017
Jev by @typesafeai is now on OpenRouter, in beta.
Jev is a System One model. Instead of generating text, it takes your app's state plus a typed question and returns a typed decision with a probability attached. There is no JSON prompting, parsing layer, and nothing to validate against.
145
284
3,877
565,014
Pin a model and copy its ID in one click from the All Models list, without leaving search. Pins carry over to the Models page, where you can pin models directly and filter down to just yours.
1
3
3,304
Search on OpenRouter got an upgrade
Hit ⌘K anywhere on openrouter.ai to search docs alongside models, providers, and pages. Before you type, browse your Pinned models plus what's Popular and Trending this week.
11
5
68
11,031
Watch the full New Defaults episode: piped.video/PS6gvq1r6IM
2
10
5,703
"The most important thing about the intelligence layer is the model.”
But what happens when developers can choose from hundreds of them?
@vincentweisser of @PrimeIntellect joined us for New Defaults to talk open models, model choice, routing, economics, and what happens as intelligence becomes increasingly open and competitive.
12
7
119
17,132
1/ Any model on OpenRouter can now run code in a hosted Linux container.
Add one tool to a Responses API request. The model writes the script, runs it, and reads back the result. Nothing to install, nothing to host.
Meet openrouter:shell.
To get started:
"tools": [{ "type": "openrouter:shell", "parameters": { "engine": "openrouter" } }]
35
27
333
39,312
Three key open-weight labs have seen monthly dollars spent on their models grow by 10x or more so far in 2026.
The closed models have seen comparatively slower growth in spending, though from much higher starting points.
25
42
421
55,696
New: US/Europe flags on providers that support geofencing requests 🇺🇸🇪🇺
openrouter.ai/docs/guides/fe…
12
4
106
14,979
OpenRouter users spent more on OpenAI models than on Anthropic models last week.
This hasn't happened for more than 2.5 years
218
412
4,423
754,115
Easily use the latest version of each OpenAI celestial body: openrouter.ai/models?q=lates…
With Sol still at 50% off
8
3
78
12,059