Compare a preset workload
Three profiles state their token bands in the open. They are presets, not traces. Cost uses the published C_1k formula. Self-hosted rows stay empty until you enter a measured tokens/s.
Catalog version 1 is frozen in this link. A later price change will not move it.
Assumptions
preset, not a measurementrevision 1- One model step per run, role reply.
- Input tokens 600 / 1,800 / 4,000 (low / preset point / high).
- Output tokens 150 / 500 / 1,200.
- Cache hit rate 0.50 on input. Retry rate 0.02.
- 2,000 runs per day at 1×. Peak concurrency 20. Burst factor 4. Peak window 6 hours.
Quality benchmark for this preset: no eval yet
Cost versus quality
Frontier draws only when a cited score exists.
Cost per 1,000 runs, lowest point estimates
DeepInfra DeepSeek-V4-Flash
$0.1909
Mistral AI Ministral 3 8B
$0.228
GroqCloud GPT OSS 20B
$0.2907
DeepInfra Llama-3.3-70B-Instruct-Turbo
$0.3468
Anthropic Claude API Claude Haiku 5.5 (prompts ≤100k)
$0.356
OpenAI API GPT-6 Luna (short context)
$0.356
DeepSeek API deepseek-flash off-peak
$0.4465
Mistral AI Mistral Small 4
$0.4575
Fireworks AI GPT OSS 120B standard
$0.4575
OpenRouter openai/gpt-oss-120b
$0.5126
Together AI gpt-oss-120B
$0.5814
GroqCloud GPT OSS 120B
$0.5814
Bar is the low-to-high token band. Tick is the preset point. Quality is not plotted.
1.0× · 2,000 runs/day
- No dedicated or provisioned GPU can be compared until you enter tokens/s and a GPU has a published hourly price.
51 options pass the filters. 2 hidden. Unpublished region, compliance, latency, or cost fails a filter that asks for it.
The link stores the preset inputs and catalog version.
Pin and compare
Up to 4. Migration effort and lock-in are a rubric, not a measurement.
Nothing pinned. Use Pin on a row below.
| Option | Cost / 1k | Quality | Source | |
|---|---|---|---|---|
Anthropic Claude API · Claude Opus 5.5 serverless_api · per_token | $4.35 – $33.05 · $14.06 | no eval yet | source checked 2026-10-20 | |
Anthropic Claude API · Claude Sonnet 5.5 serverless_api · per_token | $2.17 – $16.52 · $7.03 | no eval yet | source checked 2026-10-20 | |
Anthropic Claude API · Claude Haiku 5.5 (prompts ≤100k) serverless_api · per_token | $0.1102 – $0.8364 · $0.356 | no eval yet | source checked 2026-10-20 | |
OpenAI API · GPT-6 Astra (short context) serverless_api · per_token | $11.02 – $83.64 · $35.60 | no eval yet | source checked 2026-10-20 | |
OpenAI API · GPT-6.1 Sol (short context) serverless_api · per_token | $2.17 – $16.52 · $7.03 | no eval yet | source checked 2026-10-20 | |
OpenAI API · GPT-6 Luna (short context) serverless_api · per_token | $0.1102 – $0.8364 · $0.356 | no eval yet | source checked 2026-10-20 | |
Google Gemini API · Gemini 4 Argon (through 30 Nov 2026) serverless_api · per_token | $2.17 – $16.52 · $7.03 | no eval yet | source checked 2026-10-20 | |
Google Gemini API · Gemini 3.8 Flash (through 31 Dec 2026) serverless_api · per_token | $0.8262 – $6.27 · $2.67 | no eval yet | source checked 2026-10-20 | |
Google Gemini API · Gemini 3.1 Flash-Lite serverless_api · per_token | $0.3137 – $2.40 · $1.02 | no eval yet | source checked 2026-10-20 | |
xAI Grok API · Grok 4.7 (prompt under 200k) serverless_api · per_token | $1.68 – $12.44 · $5.35 | no eval yet | source checked 2026-10-20 | |
Mistral AI · Mistral Large 4 serverless_api · per_token · open-weight | $1.10 – $8.18 · $3.51 | no eval yet | source checked 2026-10-20 | |
Mistral AI · Mistral Small 4 serverless_api · per_token · open-weight | $0.1423 – $1.07 · $0.4575 | no eval yet | source checked 2026-10-20 | |
Mistral AI · Ministral 3 8B serverless_api · per_token · open-weight | $0.0734 – $0.5202 · $0.228 | no eval yet | source checked 2026-10-20 | |
DeepSeek API · deepseek-flash off-peak serverless_api · per_token | $0.1386 – $1.05 · $0.4465 | no eval yet | source checked 2026-10-20 | |
DeepSeek API · deepseek-flash peak serverless_api · per_token | $0.2772 – $2.09 · $0.8929 | no eval yet | source checked 2026-10-20 | |
DeepSeek API · deepseek-v4-pro off-peak serverless_api · per_token | $0.5116 – $3.81 · $1.64 | no eval yet | source checked 2026-10-20 | |
Cohere · Command (legacy, existing customers) serverless_api · per_token | $0.918 – $6.53 · $2.86 Cached price is not published, so cache hits are billed as ordinary input. | no eval yet | source checked 2026-10-20 | |
AWS Bedrock · DeepSeek v3.2 on-demand (US East/West) serverless_api · per_token | $0.6625 – $4.79 · $2.08 Cached price is not published, so cache hits are billed as ordinary input. | no eval yet | source checked 2026-10-20 | |
AWS Bedrock · Cohere Command provisioned, no commitment dedicated · per_token | — Token price is not published, so cost per 1,000 runs is empty. | no eval yet | source checked 2026-10-20 | |
Google Vertex AI · Gemini 4 Argon global (through 30 Nov 2026) serverless_api · per_token | $2.17 – $16.52 · $7.03 | no eval yet | source checked 2026-10-20 | |
Google Vertex AI · Gemini 3.8 Flash global (through 31 Dec 2026) serverless_api · per_token | $0.8262 – $6.27 · $2.67 | no eval yet | source checked 2026-10-20 | |
Together AI · DeepSeek V4.1 Flash serverless_api · per_token | $0.2772 – $2.09 · $0.8929 | no eval yet | source checked 2026-10-20 | |
Together AI · gpt-oss-120B serverless_api · per_token · open-weight | $0.1836 – $1.35 · $0.5814 Cached price is not published, so cache hits are billed as ordinary input. | no eval yet | source checked 2026-10-20 | |
Together AI · Llama 3.3 70B serverless_api · per_token | $0.7956 – $5.52 · $2.44 Cached price is not published, so cache hits are billed as ordinary input. | no eval yet | source checked 2026-10-20 | |
Fireworks AI · GPT OSS 120B standard serverless_api · per_token · open-weight | $0.1423 – $1.07 · $0.4575 | no eval yet | source checked 2026-10-20 | |
Fireworks AI · DeepSeek V4.1 Flash standard serverless_api · per_token | $0.2772 – $2.09 · $0.8929 | no eval yet | source checked 2026-10-20 | |
GroqCloud · GPT OSS 120B serverless_api · per_token · open-weight | $0.1836 – $1.35 · $0.5814 Cached price is not published, so cache hits are billed as ordinary input. | no eval yet | source checked 2026-10-20 | |
GroqCloud · GPT OSS 20B serverless_api · per_token | $0.0918 – $0.6732 · $0.2907 Cached price is not published, so cache hits are billed as ordinary input. | no eval yet | source checked 2026-10-20 | |
GroqCloud · Llama 3.3 70B serverless_api · per_token | — Token price is not published, so cost per 1,000 runs is empty. | no eval yet | source checked 2026-10-20 | |
SambaNova Cloud · gpt-oss-120b serverless_api · per_token · open-weight | $0.2249 – $1.62 · $0.7048 Cached price is not published, so cache hits are billed as ordinary input. | no eval yet | source checked 2026-10-20 | |
SambaNova Cloud · Meta-Llama-3.3-70B-Instruct serverless_api · per_token | $0.5508 – $3.92 · $1.71 Cached price is not published, so cache hits are billed as ordinary input. | no eval yet | source checked 2026-10-20 | |
DeepInfra · DeepSeek-V4-Flash serverless_api · per_token | $0.0606 – $0.4406 · $0.1909 | no eval yet | source checked 2026-10-20 | |
DeepInfra · Llama-3.3-70B-Instruct-Turbo serverless_api · per_token | $0.1102 – $0.7997 · $0.3468 Cached price is not published, so cache hits are billed as ordinary input. | no eval yet | source checked 2026-10-20 | |
OpenRouter · anthropic/claude-sonnet-5.5 serverless_api · per_token | $2.17 – $16.52 · $7.03 | no eval yet | source checked 2026-10-20 | |
OpenRouter · openai/gpt-oss-120b serverless_api · per_token · open-weight | $0.1607 – $1.19 · $0.5126 | no eval yet | source checked 2026-10-20 | |
OpenRouter · x-ai/grok-4.7 serverless_api · per_token | $1.68 – $12.44 · $5.35 | no eval yet | source checked 2026-10-20 | |
Modal · Nvidia H100 SXM5 self_hosted · per_second | — No published tokens/s for this GPU, and no measured tokens/s was entered. Cost is left empty rather than estimated. | no eval yet | source checked 2026-10-20 | |
Modal · Nvidia H200 SXM self_hosted · per_second | — No published tokens/s for this GPU, and no measured tokens/s was entered. Cost is left empty rather than estimated. | no eval yet | source checked 2026-10-20 | |
Modal · Nvidia B200 self_hosted · per_second | — No published tokens/s for this GPU, and no measured tokens/s was entered. Cost is left empty rather than estimated. | no eval yet | source checked 2026-10-20 | |
Lambda · H100 SXM 8× instance self_hosted · provisioned | — No published tokens/s for this GPU, and no measured tokens/s was entered. Cost is left empty rather than estimated. | no eval yet | source checked 2026-10-20 | |
Lambda · B200 SXM6 8× instance self_hosted · provisioned | — No published tokens/s for this GPU, and no measured tokens/s was entered. Cost is left empty rather than estimated. | no eval yet | source checked 2026-10-20 | |
Nebius · NVIDIA HGX H100 self_hosted · provisioned | — No published tokens/s for this GPU, and no measured tokens/s was entered. Cost is left empty rather than estimated. | no eval yet | source checked 2026-10-20 | |
Nebius · NVIDIA HGX H200 self_hosted · provisioned | — No published tokens/s for this GPU, and no measured tokens/s was entered. Cost is left empty rather than estimated. | no eval yet | source checked 2026-10-20 | |
Nebius · NVIDIA HGX B200 self_hosted · provisioned | — No published tokens/s for this GPU, and no measured tokens/s was entered. Cost is left empty rather than estimated. | no eval yet | source checked 2026-10-20 | |
Crusoe · H100 80GB HGX instance self_hosted · provisioned | — No published tokens/s for this GPU, and no measured tokens/s was entered. Cost is left empty rather than estimated. | no eval yet | source checked 2026-10-20 | |
Crusoe · H100 self-serve deployment self_hosted · provisioned | — No published tokens/s for this GPU, and no measured tokens/s was entered. Cost is left empty rather than estimated. | no eval yet | source checked 2026-10-20 | |
RunPod · H100 SXM pod self_hosted · provisioned | — No published tokens/s for this GPU, and no measured tokens/s was entered. Cost is left empty rather than estimated. | no eval yet | source checked 2026-10-20 | |
Together AI · HGX H100 cluster self_hosted · provisioned | — No published tokens/s for this GPU, and no measured tokens/s was entered. Cost is left empty rather than estimated. | no eval yet | source checked 2026-10-20 | |
Together AI · HGX H100 dedicated inference self_hosted · provisioned | — No published tokens/s for this GPU, and no measured tokens/s was entered. Cost is left empty rather than estimated. | no eval yet | source checked 2026-10-20 | |
Fireworks AI · H100 80 GB on-demand self_hosted · per_second | — No published tokens/s for this GPU, and no measured tokens/s was entered. Cost is left empty rather than estimated. | no eval yet | source checked 2026-10-20 | |
DeepInfra · Dedicated H100 80GB self_hosted · provisioned | — No published tokens/s for this GPU, and no measured tokens/s was entered. Cost is left empty rather than estimated. | no eval yet | source checked 2026-10-20 |