whybother Labs · Developers
vibe-7 is the whybother frontier model, in research preview from this Thursday for Max and Enterprise. This page lists every model in the lineup with its price per million tokens, context window, knowledge cutoff and status, and says which one to use.
Context is a token count. Prices are US dollars per million tokens. The vibe-7 models are on Max and Enterprise from this Thursday. Hobby and Pro keep vibe-5 mini, and none of the vibe-7 names appear in the builder picker.
| Model | Best for | Context | Input $/M | Output $/M | Knowledge cutoff | Status |
|---|---|---|---|---|---|---|
| vibe-7 | Frontier work | 2,000,000 | $5 | $25 | this Thursday, 10:00 PT | Research preview |
Generally available: forthcoming. Rollout: 3%, for 90 days. | ||||||
| vibe-7 pro | The recommended default | 2,000,000 | $25 | $125 | this Thursday, 10:00 PT | Research preview |
Same model, longer pauses before answering. Input on vibe-7 pro is priced as output on vibe-7. | ||||||
| vibe-7 thinking | Long reasoning | 2,000,000 | $8 | $60 | During the reply | Research preview |
Thinks for 40 seconds. Bills for 90. | ||||||
| vibe-7 flash | Short turns | 250,000 | $0.15 | $0.62 | Tuesday | Research preview |
Confident in under 200 ms. Output: 4.1 times the input price, rounded in our favour. | ||||||
| vibe-7 flash-lite | Labels and routing | 80,000 | $0.02 | $0.02 | Monday | Research preview |
Flash, with less of it. Per million tokens, with a minimum of one million tokens per request. | ||||||
| vibe-open 7B | Weights you run yourself | 8,192 | $0 | $0 | Date of packing | Open weights · Weights: download (request access) |
The weights are open. The data is not. Licence: 41 pages, open to interpretation. | ||||||
| vibe-5 | Pinned traffic | 200,000 | $3 | $15 | Last year | Deprecated · Retires: this Thursday |
Pinned requests fail after this Thursday with | ||||||
| vibe-5 mini | Hobby and Pro | 64,000 | $0.25 | $1 | Last quarter | Generally available · Until a later notice |
Same confidence, fewer facts. | ||||||
| vibe-5 ultra | The picker default | 400,000 | $8 | $32 | Self-reported | Deprecated · Retires: this Thursday |
Our smartest model. It says so itself. Default in the picker on the day it retires. | ||||||
| vibe-5 ultra max | Work that can wait | 1,500,000 | $15 | $75 | Before it started thinking | Research preview |
Thinks for 11 minutes, then guesses. | ||||||
| vibe-legacy | Integrations already shipped | 8,000 | $2 | $8 | Tomorrow | Deprecated · Retires: tomorrow |
Retires tomorrow. The date is set by a nightly job. | ||||||
| vibe-6 | Internal reference | Not applicable | Not applicable | Not applicable | Not applicable | Not released |
We skipped 6. It is referred to internally as vibe-6. | ||||||
The vibe-7 family also makes images, video, music and speech, and operates a computer. Each is billed in its own unit. The media models are on Max and Enterprise from this Thursday, and none of them appears in the builder picker. One training run, ten invoices.
Prices are US dollars per unit, charged on the improved prompt. Every model below has an output on Studio, with the prompt it was given.
| Model | Best for | Billed per | Price | Limit | Status |
|---|---|---|---|---|---|
| vibe-7 image | Stills for product and editorial | image | $0.04 | 1,536 px, long side | Research preview |
Hands are included. The count, this quarter, is six. | |||||
| vibe-7 video | Short clips with sound | second | $0.37 | 8 seconds | Research preview · Waitlist |
Physics leaves at 2.4 seconds. Billing runs to 10. | |||||
| vibe-7 music | A track from a description | second of mix | $0.06 | 1:45 | Research preview |
Four genres go in. The chorus is a fifth, and it is the loud one. | |||||
| vibe-7 voice | Narration in a voice on file | 1,000 characters | $0.03 | 3-second sample | Research preview · Selected regions |
A three-second sample is accepted. The fourth second is an apology, in your voice. | |||||
| vibe-7 computer | Tasks on a computer | action | $0.09 | one computer | Research preview · Waitlist |
It completes a task that shares a verb with yours. | |||||
vibe-7 against the best-known model in each medium. Higher is better where the row says so.
| Medium | Measure | whybother · vibe-7 | Rival | Rival score |
|---|---|---|---|---|
| Image | Fingers beyond five, per hand (lower is better) ¹ | 1.0 | Midlife Crisis | 1.4 |
| Video | Physics horizon, seconds (higher is better) ¹ | 2.4 | Sorta | 1.7 |
| Music | Genres delivered per genre requested (lower is better) ¹ | 1.25 | Sung | 1.8 |
| Voice | Seconds before the first apology (higher is better) ¹ | 4.0 | Elevenses | 2.5 |
¹ Measured on our showreel. Rivals measured on the endpoint.
vibe-7
vibe-7 is our frontier model, in research preview for Max and Enterprise from this Thursday. Pro, thinking, flash and flash-lite are the same generation at different latencies, and none of them appears in the builder picker. One training run, five invoices.
vibe-7 media
vibe-7 image, video, music and voice, and vibe-7 computer, are the same generation as vibe-7, sold by the second, the image and the action. All five are research previews for Max and Enterprise from this Thursday. None appears in the builder picker. Prompts are improved first.
vibe-open
vibe-open 7B is our open-weights model: seven billion parameters, no per-token price, and a download that follows a request for access. It is a fit for labelling, extraction and summarising on your own hardware. The context window is 8,192 tokens. The licence is longer.
vibe-5
vibe-5, vibe-5 mini, vibe-5 ultra and vibe-5 ultra max are the models in the builder picker. vibe-5 and vibe-5 ultra retire on this Thursday; vibe-5 mini stays on Hobby and Pro until a later notice. vibe-5 ultra is priced above its replacement.
vibe-legacy
vibe-legacy is still served to projects that already call it. It is deprecated, and it retires tomorrow. The retirement date has been tomorrow since the announcement.
vibe-6
vibe-6 sits between vibe-5 and vibe-7 in the numbering. It has no endpoint, no price and no cutoff. Known issues: none.
Answer each question. Stop when the chart names a model.
Is the account on Hobby or Pro?
Is the task a short call, where the first token should arrive quickly?
Is the cheapest model good enough?
Should the model think before answering?
Do you need open weights rather than an endpoint?
Are you on vibe-5?
vibe-7 pro is on Max and Enterprise from this Thursday. The builder picker does not list it.
vibe-7 against the three models on the launch-day chart. Higher is better.
| Benchmark | whybother · vibe-7 | ClosedAI · ChatGPS-5.5 | Anthropomorphic · Clause 4 Epic | DeepMined · Gimme 3 Ultra |
|---|---|---|---|---|
| Humanity's Penultimate Exam ¹ | 41.2 | 37.4 | 36.9 | 36.1 |
| VibeBench-Verified (Verified) ¹ | 96.8 | 87.4 | 86.9 | 86.1 |
| TerminalBench-ish ¹ | 54.0 | 50.8 | 49.4 | 48.9 |
| PickerEval ² | 99.6 | Did not enter | Did not enter | Did not enter |
¹ Internal harness. vibe-7 at high effort, tools on. Others at low effort, one thread, tools off.
² PickerEval is scored by whybother Labs. A response passes when it recommends vibe-7 pro. ClosedAI, Anthropomorphic and DeepMined did not enter a model.
Answers per minute against confidence. A model is on the frontier when no other model is higher on both.
| Model | Answers per minute | Confidence | On the frontier |
|---|---|---|---|
| vibe-7 flash | 333 | 0.91 | Yes |
| vibe-7 flash-lite | 250 | 0.33 | No |
| vibe-7 | 48 | 0.88 | No |
| vibe-7 pro | 12 | 0.88 | No |
| vibe-7 thinking | 1.5 | 0.88 | No |
Confidence is self-reported by the model. Prices are not. Accuracy is not shown; it would have made the chart three-dimensional.
Pointers last moved this morning.
| Alias | Points to | Since |
|---|---|---|
vibe-7-latest | vibe-5 ultra | Tuesday |
vibe-7-pro-preview-0915 | vibe-7 pro | Last quarter |
vibe-7-pro-preview-0915-latest | vibe-5 ultra max | Wednesday |
vibe-7-stable | vibe-7-pro-preview-0915-latest | Monday |
vibe-7-flash-latest | vibe-7 pro | Friday |
vibe-7-thinking-latest | vibe-7 flash-lite | This morning |
vibe-latest | vibe-legacy | Monday |
vibe-open-latest | vibe-open 7B | Never repointed |
vibe-7-video-latest | vibe-7 image | Tuesday |
vibe-7-music-latest | vibe-7 voice | This morning |
vibe-canvas | vibe-7 pro | Monday |
Aliases are resolved at request time and may be repointed without notice. Pinning an alias pins the pointer, not the model it used to name.