Qwen3.8 2.4T A95B API pricing

5 providers serve Qwen3.8 2.4T A95B. Qwen3.8 2.4T A95B is an open-weight model (Qwen3.8-Max) with 2400B total parameters, 95B active per token (mixture-of-experts), up to 1049K context. Prices below are per 1M tokens, cheapest input first.

Refreshed 2026-11-22 · list prices from provider APIs
Input floor $2.00/1M at Together AI, Streamlake, DeepInfra and 2 more Output floor $6.00/1M at Together AI, Streamlake, DeepInfra and 2 more Find your workload winner →
Provider Input $/1M Output $/1M Cache read $/1M Context
Together AI $2.00 $6.00output floor $0.250 1010K Use
Streamlake $2.00 $6.00 $0.250 984K
DeepInfra $2.00 $6.00 $0.200 262K Use
Aihubmix $2.00 $6.00 $0.500 1000K
OpenRouter marketplace quote $2.00 $6.00 $0.250 1049K Use

The cheapest way to run Qwen3.8 2.4T A95B

On an illustrative 70M-input / 30M-output monthly workload, today's lowest listed cost is roughly $320/month — the same at Together AI, Streamlake, DeepInfra and 2 more. Input-heavy, output-heavy and cache-heavy workloads can produce different winners. Use the live mix calculator below rather than combining floors from two different providers. On output-token cost alone, the API floor remains below our self-hosting estimate of ~$11.71/1M. Details below. Run your own numbers in the breakeven calculator, or read the breakeven math.

Your actual monthly cost

Uses today's list prices

Cache read uses the listed cache rate where available; otherwise it falls back to normal input price. Batch, write-cache, volume and negotiated discounts are excluded.

Or run it yourself: Qwen3.8 2.4T A95B self-hosting economics

Qwen3.8 2.4T A95B is open-weight (Qwen3.8-Max), so the API price above competes with the GPU-hour market. 2400B total parameters (MoE, ~95B active per token) needs roughly 3300 GB VRAM at FP8 or 1650 GB at INT4, KV-cache headroom included.

No single GPU fits Qwen3.8 2.4T A95B at FP8 (3300 GB needed), so it needs a multi-GPU node. Cheapest tracked option: 18× MI300X (192 GB each) at roughly $9.00/hr total (RunPod). Rent 18× MI300X at RunPod → INT4 quantization drops the requirement to 1650 GB. Tune it in the calculator.
Breakeven estimate: a well-batched vLLM deployment on 18× MI300X ($9.00/hr) at ~427 aggregate tok/s and 50% utilization works out to roughly $11.71 per 1M output tokens, versus $6.00 via the cheapest output-token API (Together AI, Streamlake, DeepInfra and 2 more). The API is cheaper unless your utilization or throughput beats these assumptions. Planning estimate: throughput varies with hardware, quantization, batch size and context. Tune it in the calculator.

Related models

ModelCheapest in $/1MCheapest out $/1MProviders
Qwen3.6 35B A3B $0.050 $0.400 13
Qwen3.8 Max (0902) $1.65 $4.95 12
Qwen3.8 27B $0.035 $0.500 12
Qwen3 Next 80B A3B Instruct $0.090 $0.900 11
Qwen3 235B A22B $0.140 $0.540 11
Qwen3.5 397B A17B $0.164 $0.986 11

FAQ

What is the cheapest API for Qwen3.8 2.4T A95B?

As of 2026-11-22, the lowest input price for Qwen3.8 2.4T A95B, $2.00 per 1M tokens, is the same list price at Together AI, Streamlake, DeepInfra and 2 more. The lowest output price, $6.00 per 1M tokens, is likewise shared by the same providers. The cheapest provider for a real workload depends on its input/output mix.

How much VRAM do you need to self-host Qwen3.8 2.4T A95B?

Qwen3.8 2.4T A95B has 2400B parameters, so plan for roughly 3300 GB of VRAM at FP8 or 1650 GB at INT4/AWQ, KV-cache headroom included. That exceeds a single GPU. A typical node is 18× MI300X.

← All models · GPU rental prices · Breakeven calculator · Get this data as JSON