Thinking Machines · Chat and reasoning · 952B parameters · 41B active · apache-2.0
Inkling
Needs 1,143 GB at FP8 / INT8. The smallest real configuration today is 8× MI300X at TensorWave, $13.68 per hour, about $9,986 a month at 730 hours.
Parameters952B41B active per token
Context—maximum published
Checkpoint1,905 GBBF16, as published
Licenseapache-2.0check before commercial use
ReleasedJul 15, 2026documented date
Smallest configuration · FP8 / INT8
8× MI300X1536 GB VRAM · 74% used
- Per hour
- $13.68
- Per month
- $9,986
- Cheapest at
TensorWave
Memory · 3 precisionsHow much VRAM it needs
Weights plus runtime overhead. The context dimension appears once the model's architecture is on file.
| Precision | Weights | Weights + headroom |
|---|---|---|
| BF16published checkpoint | 1,905 GB | 2,286 GB |
| FP8 / INT8estimated | 952 GB | 1,143 GB |
| INT4 / MXFP4estimated | 571 GB | 686 GB |
Cross-checked on Hugging Face Oct 28, 2026: 952.4B parameters in the safetensors headers, 1904.8 GB of weight files, attention layout from config.json (66 layers, 8 KV heads × 128). Method: weights as published at the native precision, verified on Hugging Face 2026-10-28; other precisions estimated at 1 (8-bit) and 0.6 (4-bit) bytes per parameter; × 1.2 runtime headroom; KV cache = 4096 bytes per token per layer × 66 full-attention layers (FP16 cache, one sequence).config.json ↗
Smallest configuration per GPU · 110 GPUs · cheapest eligible on-demand offer per GPU count, every provider on fileWhat it costs to run, GPU by GPU
The fewest GPUs of one type whose total memory fits the model at short context, priced from the cheapest eligible on-demand offer for exactly that count (45-day window, newest reading per provider), with the next providers beside it. Monthly = 730 hours.
| GPU | Smallest config | VRAM used | $ / hour | $ / month | Cheapest at |
|---|---|---|---|---|---|
| MI300XBest fit | 8× 192 GB1536 GB total | 74% | $13.68 | $9,986 | TensorWaveTrust 64 · read Nov 1 · receipt · also Cyfuture AI $22.38, AtmosCompute $26.77 |
| MI325X | 8× 256 GB2048 GB total | 56% | $18.00 | $13,140 | TensorWaveTrust 64 · read Nov 1 · receipt · also Cyfuture AI $24.34 |
| B300 | 4× 288 GB1152 GB total | 99% tight | $23.29 | $17,004 |  UpCloudTrust 75 · read Nov 1 · receipt · also FPT AI Factory $27.96, Vast.ai $37.50 |
| MI355X | 8× 288 GB2304 GB total | 50% | $23.60 | $17,228 | TensorWaveTrust 64 · read Nov 1 · receipt |
| B200 | 8× 180 GB1440 GB total | 79% | $32.91 | $24,026 | UpCloudTrust 75 · read Nov 1 · receipt · also Airon AI $33.04, Deep Infra $39.12 |
| B300 CC | 4× 288 GB1152 GB total | 99% tight | $46.31 | $33,806 |  VerdaTrust 81 · read Nov 1 · receipt |
| GB300 | 4× 288 GB1152 GB total | 99% tight | $52.21 | $38,113 | VerdaTrust 81 · read Nov 1 · receipt |
| B200 CC | 8× 180 GB1440 GB total | 79% | $72.37 | $52,830 | VerdaTrust 81 · read Nov 1 · receipt |
Throughput estimates · Oct 4, 2026Tokens per second, and what a million costs
Chat and reasoning. Speed estimated from memory bandwidth and model size, never measured. The cost per 1M output tokens is an estimate at the prices read on Oct 4, 2026; it is not recomputed here.
| GPU | Cards | Precision | tok/s per user | per 1M output tokens · estimate |
|---|---|---|---|---|
| MI300X | 8 | FP8 | 30.8 | $4.91 |
| B200 | 8 | FP8 | 53.6 | $6.44 |
| H200 | 8 | FP8 | 39.8 | $7.27 |
cards, precision and speed estimated from memory bandwidth and model size; cost recomputed by the site from verified prices · cheapest per output at the prices of Oct 4, 2026: MI300XWhy estimates, and their limits →
Inkling · related