LLMBOTTLENECK.COM
Aleph-Alpha / kolibri1

Kolibri-1-BF16

78.1 billion parameters, routing 6 of 384 experts per token.

Open in the calculator →

Architecturepublished data
Quick answer
Kolibri-1-BF16 needs about 48.5 GB at Q4_K_M — the format most people download — with 8,192 tokens of context. It fits entirely on 7 of 18 common devices, starting with the Apple M4 Pro (64 GB). What else fits in 64 GB. No hardware for it? Rent a GPU that runs it, priced live.

Sized at 8,192 tokens of context with the whole model in device memory, across 18 common devices. Speeds are estimates, not benchmarks.

Find the Apple M4 Pro, the smallest card listed here that holds it: Amazon ↗ · eBay (new and used) ↗Store links may pay us a commission. They never decide which card is suggested — the memory arithmetic does.

Aleph Alpha's own release (Apache-2.0), 2026-10-02: the bfloat16 checkpoint of Kolibri 1, a mixture-of-experts model whose card states 78,103,074,560 total and 3,457,573,120 active parameters per token. The companion repository Aleph-Alpha/Kolibri-1 holds the same model in FP8 and declares this repository as its base_model. Licence declared on the repository: apache-2.0.

Weight-file size by format

FormatSizeRangeBasis
FP16156.3 GB155.5 GB – 157.1 GBreconstructed size
Q8_083.1 GB82.7 GB – 83.6 GBreconstructed size
Q6_K64.2 GB63.9 GB – 64.6 GBreconstructed size
Q5_K_M55.6 GB55.3 GB – 55.9 GBreconstructed size
Q5_053.9 GB53.6 GB – 54.2 GBreconstructed size
Q4_K_M47.5 GB47.2 GB – 47.7 GBreconstructed size
Q4_044.2 GB43.9 GB – 44.4 GBreconstructed size
Q3_K_M38.8 GB36.1 GB – 41.5 GBreconstructed size
Q2_K28.5 GB27.4 GB – 29.7 GBreconstructed size

A file size is not the memory a run needs: the KV cache and the runtime reserve come on top, and the calculator adds both.

What the labels mean
published data
Read from a published source — a file's byte count, a model's configuration or a manufacturer's specification — or exact arithmetic on such values. Not a measurement on a machine.
reconstructed size
Weight size reconstructed from the pinned architecture, because no published file exists.
size range
Only a lower and an upper bound are claimed for this weight size.

Hardware ladder

Q4_K_M at 8,192 tokens · 7 of 18 devices hold it entirely
DeviceMemoryVerdictDecodeCalculator
NVIDIA GeForce RTX 507048.5 GB of 12 GB39 layers on system RAM~32 tok/s with offloadOpen →
AMD Radeon™ RX 9070 XT48.5 GB of 16 GB35 layers on system RAM~35 tok/s with offloadOpen →
NVIDIA GeForce RTX 408048.5 GB of 16 GB35 layers on system RAM~35 tok/s with offloadOpen →
NVIDIA GeForce RTX 508048.5 GB of 16 GB35 layers on system RAM~36 tok/s with offloadOpen →
NVIDIA GeForce RTX 5070 Ti48.5 GB of 16 GB35 layers on system RAM~36 tok/s with offloadOpen →
NVIDIA GeForce RTX 5060 Ti 16GB48.5 GB of 16 GB35 layers on system RAM~34 tok/s with offloadOpen →
NVIDIA GeForce RTX 409048.5 GB of 24 GB26 layers on system RAM~46 tok/s with offloadOpen →
AMD Radeon™ RX 7900 XTX48.5 GB of 24 GB26 layers on system RAM~47 tok/s with offloadOpen →
NVIDIA GeForce RTX 309048.5 GB of 24 GB26 layers on system RAM~46 tok/s with offloadOpen →
NVIDIA GeForce RTX 509048.5 GB of 32 GB18 layers on system RAM~67 tok/s with offloadOpen →
NVIDIA RTX 6000 Ada Generation48.5 GB of 48 GB1 layers on system RAM~223 tok/s with offloadOpen →
Apple M4 Pro48.5 GB of 64 GBfits~65 tok/sOpen →
Apple M5 Pro48.5 GB of 64 GBfits~70 tok/sOpen →
AMD Ryzen AI Max+ 395 with Radeon 8060S48.5 GB of 128 GBfits~80 tok/sOpen →
Apple M3 Max48.5 GB of 128 GBfits~81 tok/sOpen →
Apple M4 Max48.5 GB of 128 GBfits~95 tok/sOpen →
NVIDIA DGX Spark48.5 GB of 128 GBfits~75 tok/sOpen →
Apple M2 Ultra48.5 GB of 192 GBfits~111 tok/sOpen →

Decode is a calibrated estimate from the published calibration; capacity is the manufacturer's published ceiling, not guaranteed free memory. Fitting in memory is not the same as loading: whether the runtime and version you have supports this architecture and format on that machine has not been tested here. Offloaded rows assume there is enough system RAM for the overflow — the calculator checks that against the RAM you declare. Each “Open” link keeps this model, format and context.

Run it in the cloud

Rent a GPU that runs Kolibri-1-BF16

Every rentable machine that holds the whole model, from Vast.ai and RunPod, with this site's speed estimate for each and the provider's own price, read live. Cheapest first; sort by cost per token or speed instead, or switch the format.

Prices · 18:30 UTC, 22 Sept
19 configurations in stock · 14 out of stock
MachineSpeedPer hourPer M tokensWhereRent
4× RTX 5060 Ti 16GB64 GBcheapest~491 tok/sFaster than you read$0.51$0.29Vast.aiMarketplace · 99.5% reliableRent on Vast.ai
2× RTX 509064 GBrecommendedbest value~983 tok/sFaster than you read$0.54$0.15Vast.aiMarketplace · 99.6% reliableRent on Vast.ai
8× RTX 306096 GB~728 tok/sFaster than you read$0.59$0.22Vast.aiMarketplace · 99.7% reliableRent on Vast.ai
4× RTX 309096 GB~1026 tok/sFaster than you read$0.60$0.16Vast.aiMarketplace · 99.4% reliableRent on Vast.ai
4× RTX 4060 Ti 16GB64 GB~316 tok/sFaster than you read$0.80$0.71Vast.aiMarketplace · 99.3% reliableRent on Vast.ai
2× RTX 6000 Ada96 GB~526 tok/sFaster than you read$1.07$0.57Vast.aiMarketplace · 99.4% reliableRent on Vast.ai
1× RTX PRO 600096 GB~491 tok/sFaster than you read$1.14$0.64Vast.aiMarketplace · 98.8% reliableRunPod $1.69/h Rent on Vast.ai
1× A100 80GB PCIe80 GB~530 tok/sFaster than you read$1.19$0.62RunPodCommunity CloudRent on RunPod

First time renting a GPU? How it works, in four steps
  1. Create an account and add credit. Both providers are prepaid: $10 is enough to try any card on this page for hours. A new RunPod account that arrives through a referral link gets a one-time $5 credit when it first adds $10.
  2. Pick the card and the count shown here, and a template: llama.cpp or vLLM for an OpenAI-compatible API, or one with a web chat if you only want to talk to the model. “How to launch it” below gives the exact image and command.
  3. Wait for the download. The weights are fetched on the machine; tens of gigabytes take a few minutes on a data-centre connection. You pay by the second from the moment it starts.
  4. Stop it when you are done. A running machine bills even when idle. A stopped one costs nothing per hour, though its disk is usually still billed until you delete it.
How to launch it

No Q4_K_M file of Kolibri-1-BF16 is catalogued here, so there is no exact command to vouch for. Start the machine from the llama.cpp server image ghcr.io/ggml-org/llama.cpp:server-cuda and point -hf at a Q4_K_M build from the publisher or a quantizer you trust — search Hugging Face for one. Check its file size against the memory figure on this page before you rent.

Every machine holds the whole model at Q4_K_M and 8,192 tokens of context, no offload. Speeds are this site’s single-stream decode estimates; prices are what each provider’s own API quoted, on-demand, for the whole machine. Vast hosts below 98% measured reliability are left out.

Referral links Vast.ai, RunPod and Novita pay us a share of what you spend if you sign up through these buttons. It costs you nothing, and it never decides an order or a recommendation: both are computed from the live price and the speed, and options that pay us nothing are listed and recommended on the same terms. How we rank

Buying a card instead, or paying by the token? Compare a year of Kolibri-1-BF16 three ways →

Under the hood

Architecture, read from the publisher’s file

The numbers every figure above is computed from, with the file they came from.

Architecture

✓ Architecture read from the published config.json

Retrieved 2026-10-03 at pinned commit 7a8f290e7858; parameter count from the safetensors index at the same revision.

Architecture
kolibri1
Layers
50
Hidden size
2,560
Attention heads
48
KV heads
4
Head dimension
128
Vocabulary
128,000
Context ceiling
262,144
Sliding window
513
RoPE theta
10,000
Experts
384
Experts / token
6
Expert width
512
Shared expert width
512

Where the memory goes

tokenembedding× 50 decoder blocksGrouped-query attention48 query · 4 KV headswindow 513Routed experts6 of 384 per tokenall residentoutputprojectiongrows with contextcapacity ≠ traffic

Grouped-query attention shares each key/value head across 12 query heads, so the KV cache is 8% of what multi-head attention would need at the same context.

Open the exact official config.json ↗

Derivatives this page also answers for

Each of these repositories declares Kolibri-1-BF16 as its base, and its published configuration matches this one on every field that decides memory. Its page lists what was compared and any files it publishes. Paste any other repository.

Aleph-Alpha/Kolibri-1

Use this answer in your own product

As JSON, from the API
curl -s https://llmbottleneck.com/v1/analyze \
  -H "Authorization: Bearer $LLMB_KEY" -H "content-type: application/json" \
  -d '{"model":"aleph-alpha-kolibri-1-bf16","quantization":"Q4_K_M","context":8192,"hardware":"nvidia-geforce-rtx-4090"}'

Same engine, same evidence, every field sourced. Free key in one step →

On your page, as a widget
<script src="https://llmbottleneck.com/widget.js"
  data-model="aleph-alpha-kolibri-1-bf16" data-quantization="Q4_K_M"
  data-hardware="nvidia-geforce-rtx-4090" data-context="8192"></script>

No key needed. Unbranded, with your own buy button, on Pro and Business →

Citing this page

LLM Bottleneck. “Kolibri-1-BF16 VRAM and hardware requirements.” Architecture from Aleph-Alpha/Kolibri-1-BF16 at revision 7a8f290e7858, retrieved 2026-10-03. https://llmbottleneck.com/models/aleph-alpha-kolibri-1-bf16

Every figure above is either the published value or a reconstruction whose measured error is on the accuracy page.

llmbottleneck
catalogue 2026-10-03models 327devices 135