GPT-OSS 20B on Groq API Pricing: Blazing Speed at Budget Rates
Explore GPT-OSS 20B on Groq API pricing ($0.07/M input, $0.30/M output), ultra-high token generation speed on Groq LPUs, and budget cost efficiency.
Full specs, context window and API limits →How much does GPT-OSS 20B cost per million tokens?
GPT-OSS 20B on Groq costs $0.07 per million input tokens and $0.30 per million output tokens ($0.1275/M blended at 3:1). Delivers lightning-fast inference on Groq LPUs for classification and code generation. Verified 2026-09-08.
Verified 2026-09-07 — source
How much does GPT-OSS 20B cost per 1,000 requests?
Computed from generated token pricing. Each row assumes the listed input and output tokens per request; output is adjusted by this model's measured 2.93× verbosity factor.
| Request shape | Input tokens | Output tokens | Cost / 1,000 requests |
|---|---|---|---|
| Short | 100 | 50 | $0.0515 |
| Medium | 1,000 | 500 | $0.5145 |
| Long | 4,000 | 2,000 | $2.0580 |
Formula: ((input price × input tokens) + (output price × output tokens × verbosity factor)) ÷ 1,000,000 × 1,000. Assumptions: short 100/50, medium 1,000/500, long 4,000/2,000 input/output tokens per request. Verbosity run: 2026-06-16T20:31:30.728Z.
Three model-specific pricing decisions
Groq-hosted 20B economics stay separate from open-weight specifications and the 120B page. Every bill below uses text-token units.
1. Fixed workload bills and reasoning-output sensitivity
| Workload | Input / output | 100K requests | Expansion treatment |
|---|---|---|---|
| Classification | 300 / 30 | $3.15 | Text tokens |
| Coding agent | 4,000 / 1,000 | $60.00 | Text tokens |
| Reasoning ×2 | 4,000 / 1,000 | $90.00 | Output doubled explicitly |
2. 20B → 120B cost-per-accepted-run threshold
| Fixed input / output | GPT-OSS 20B | GPT-OSS 120B (Cerebras) | Narrow decision boundary |
|---|---|---|---|
| Coding run · 4,000 / 1,000 | $60.00 | $215.00 | 15% accepted-result uplift required |
| Agent run · 8,000 / 1,200 | $96.00 | $370.00 | 20% accepted-result uplift required |
Formula: requests × (input tokens × input $/M + output tokens × output $/M) ÷ 1,000,000. The uplift is a planning threshold, not a measured quality claim.
3. Price / TTFT / throughput frontier
| Host/model | Listed input / output per M | Speed evidence | Capacity boundary |
|---|---|---|---|
| GPT-OSS 20B | $0.07 / $0.30 | 1120 tokens/sec; TTFT 140 ms; 5 measured samples | Rate limit and concurrency unavailable |
| GPT-OSS 120B (Cerebras) | $0.35 / $0.75 | 2450 tokens/sec; TTFT 90 ms; 5 measured samples | Do not infer queue capacity |
Verified 2026-04-06. “Unavailable” means no compatible dated evidence was found; it is never treated as zero. First-party price source · Run this scenario.
All three decisions below are computed for GPT-OSS 20B; fixed inputs, formulas, dated sources, speed sample state, and unavailable mechanics are visible.
Exact-model pricing guide · verified 2026-09-07
Exact model boundary: Groq gpt-oss-20b (slug gpt-oss-20b). First-party provider pricing and API documentation remain fact owners.
LPU high-throughput token pricing and monthly spend matrix
Frozen scenario board. Formula / deterministic rule: monthly_spend = calls * ((in_tokens * 0.07 + out_tokens * 0.30) / 1M) Boundary: Owns Groq LPU token expenditure calculations for GPT-OSS 20B.
| Frozen scenario | Model, identity, provider, and evidence fields | Result | State |
|---|---|---|---|
| 50K low-latency conversational replies | model=gpt-oss-20b; slug=gpt-oss-20b; provider=Groq; scenario=50K low-latency conversational replies; workload; prompt tokens; completion tokens; monthly spend; speed advantage; cost verdict; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-07; measurement versus assumption=explicit | Unavailable — 50K low-latency conversational replies is a frozen fixture pending exact identity, configuration, and denominator joins. | UNTESTED — assumption cannot establish a verdict |
| 200K code generation tasks | model=gpt-oss-20b; slug=gpt-oss-20b; provider=Groq; scenario=200K code generation tasks; workload; prompt tokens; completion tokens; monthly spend; speed advantage; cost verdict; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-07; measurement versus assumption=explicit | Unavailable — 200K code generation tasks is a frozen fixture pending exact identity, configuration, and denominator joins. | UNTESTED — assumption cannot establish a verdict |
| 1M automated customer triage passes | model=gpt-oss-20b; slug=gpt-oss-20b; provider=Groq; scenario=1M automated customer triage passes; workload; prompt tokens; completion tokens; monthly spend; speed advantage; cost verdict; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-07; measurement versus assumption=explicit | Unavailable — 1M automated customer triage passes is a frozen fixture pending exact identity, configuration, and denominator joins. | UNTESTED — assumption cannot establish a verdict |
| high-concurrency request surge | model=gpt-oss-20b; slug=gpt-oss-20b; provider=Groq; scenario=high-concurrency request surge; workload; prompt tokens; completion tokens; monthly spend; speed advantage; cost verdict; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-07; measurement versus assumption=explicit | Unavailable — high-concurrency request surge is a frozen fixture pending exact identity, configuration, and denominator joins. | UNTESTED — assumption cannot establish a verdict |
| batch offline ingestion queue | model=gpt-oss-20b; slug=gpt-oss-20b; provider=Groq; scenario=batch offline ingestion queue; workload; prompt tokens; completion tokens; monthly spend; speed advantage; cost verdict; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-07; measurement versus assumption=explicit | Unavailable — batch offline ingestion queue is a frozen fixture pending exact identity, configuration, and denominator joins. | UNTESTED — assumption cannot establish a verdict |
| unresolved billing currency | model=gpt-oss-20b; slug=gpt-oss-20b; provider=Groq; scenario=unresolved billing currency; workload; prompt tokens; completion tokens; monthly spend; speed advantage; cost verdict; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-07; measurement versus assumption=explicit | Unavailable — unresolved billing currency has no matched, dated bilateral observation. | FAIL CLOSED — manual, probe, or source evidence required |
First-party provenance: Groq Cloud official pricing; verification date 2026-09-07. Missing or conflicting joins fail closed.
Inference turnaround latency and streaming UX SLA audit
Frozen scenario board. Formula / deterministic rule: turnaround = ttft + (tokens_out / tps); Groq LPU ultra-high generation throughput Boundary: Owns turnaround time benchmarks and user experience responsiveness.
| Frozen scenario | Model, identity, provider, and evidence fields | Result | State |
|---|---|---|---|
| real-time conversational streaming (<80ms) | model=gpt-oss-20b; slug=gpt-oss-20b; provider=Groq; scenario=real-time conversational streaming (<80ms); task; response SLA; TTFT; generation speed; SLA compliance; responsiveness champion; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-07; measurement versus assumption=explicit | Unavailable — real-time conversational streaming (<80ms) is a frozen fixture pending exact identity, configuration, and denominator joins. | UNTESTED — assumption cannot establish a verdict |
| automated code autocomplete (<150ms) | model=gpt-oss-20b; slug=gpt-oss-20b; provider=Groq; scenario=automated code autocomplete (<150ms); task; response SLA; TTFT; generation speed; SLA compliance; responsiveness champion; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-07; measurement versus assumption=explicit | Unavailable — automated code autocomplete (<150ms) is a frozen fixture pending exact identity, configuration, and denominator joins. | UNTESTED — assumption cannot establish a verdict |
| instant document summary (<300ms) | model=gpt-oss-20b; slug=gpt-oss-20b; provider=Groq; scenario=instant document summary (<300ms); task; response SLA; TTFT; generation speed; SLA compliance; responsiveness champion; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-07; measurement versus assumption=explicit | Unavailable — instant document summary (<300ms) is a frozen fixture pending exact identity, configuration, and denominator joins. | UNTESTED — assumption cannot establish a verdict |
| long-form synthetic data generation | model=gpt-oss-20b; slug=gpt-oss-20b; provider=Groq; scenario=long-form synthetic data generation; task; response SLA; TTFT; generation speed; SLA compliance; responsiveness champion; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-07; measurement versus assumption=explicit | Unavailable — long-form synthetic data generation is a frozen fixture pending exact identity, configuration, and denominator joins. | UNTESTED — assumption cannot establish a verdict |
| network contention latency buffer | model=gpt-oss-20b; slug=gpt-oss-20b; provider=Groq; scenario=network contention latency buffer; task; response SLA; TTFT; generation speed; SLA compliance; responsiveness champion; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-07; measurement versus assumption=explicit | Unavailable — network contention latency buffer is a frozen fixture pending exact identity, configuration, and denominator joins. | UNTESTED — assumption cannot establish a verdict |
| unmeasured speed fixture | model=gpt-oss-20b; slug=gpt-oss-20b; provider=Groq; scenario=unmeasured speed fixture; task; response SLA; TTFT; generation speed; SLA compliance; responsiveness champion; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-07; measurement versus assumption=explicit | Unavailable — unmeasured speed fixture is a frozen fixture pending exact identity, configuration, and denominator joins. | UNTESTED — assumption cannot establish a verdict |
First-party provenance: Groq Cloud documentation; verification date 2026-09-07. Missing or conflicting joins fail closed.
Model tier escalation and multi-model routing model
Frozen scenario board. Formula / deterministic rule: blended_cost = 20b_volume * 20b_cost + 120b_volume * 120b_cost Boundary: Owns two-tier architectural routing between fast 20B triage and deep 120B generation.
| Frozen scenario | Model, identity, provider, and evidence fields | Result | State |
|---|---|---|---|
| 100% GPT-OSS 20B baseline ($0.07/$0.30) | model=gpt-oss-20b; slug=gpt-oss-20b; provider=Groq; scenario=100% GPT-OSS 20B baseline ($0.07/$0.30); escalation mix; total monthly queries; blended spend; cost savings vs pure 120B; accuracy retention; routing recommendation; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-07; measurement versus assumption=explicit | Unavailable — 100% GPT-OSS 20B baseline ($0.07/$0.30) is a frozen fixture pending exact identity, configuration, and denominator joins. | UNTESTED — assumption cannot establish a verdict |
| 90% 20B triage / 10% 120B escalation | model=gpt-oss-20b; slug=gpt-oss-20b; provider=Groq; scenario=90% 20B triage / 10% 120B escalation; escalation mix; total monthly queries; blended spend; cost savings vs pure 120B; accuracy retention; routing recommendation; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-07; measurement versus assumption=explicit | Unavailable — 90% 20B triage / 10% 120B escalation is a frozen fixture pending exact identity, configuration, and denominator joins. | UNTESTED — assumption cannot establish a verdict |
| 80% 20B triage / 20% 120B escalation | model=gpt-oss-20b; slug=gpt-oss-20b; provider=Groq; scenario=80% 20B triage / 20% 120B escalation; escalation mix; total monthly queries; blended spend; cost savings vs pure 120B; accuracy retention; routing recommendation; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-07; measurement versus assumption=explicit | Unavailable — 80% 20B triage / 20% 120B escalation is a frozen fixture pending exact identity, configuration, and denominator joins. | UNTESTED — assumption cannot establish a verdict |
| 50% 20B triage / 50% 120B escalation | model=gpt-oss-20b; slug=gpt-oss-20b; provider=Groq; scenario=50% 20B triage / 50% 120B escalation; escalation mix; total monthly queries; blended spend; cost savings vs pure 120B; accuracy retention; routing recommendation; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-07; measurement versus assumption=explicit | Unavailable — 50% 20B triage / 50% 120B escalation is a frozen fixture pending exact identity, configuration, and denominator joins. | UNTESTED — assumption cannot establish a verdict |
| 100% direct GPT-OSS 120B execution | model=gpt-oss-20b; slug=gpt-oss-20b; provider=Groq; scenario=100% direct GPT-OSS 120B execution; escalation mix; total monthly queries; blended spend; cost savings vs pure 120B; accuracy retention; routing recommendation; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-07; measurement versus assumption=explicit | Unavailable — 100% direct GPT-OSS 120B execution is a frozen fixture pending exact identity, configuration, and denominator joins. | UNTESTED — assumption cannot establish a verdict |
| unresolved confidence threshold trigger | model=gpt-oss-20b; slug=gpt-oss-20b; provider=Groq; scenario=unresolved confidence threshold trigger; escalation mix; total monthly queries; blended spend; cost savings vs pure 120B; accuracy retention; routing recommendation; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-07; measurement versus assumption=explicit | Unavailable — unresolved confidence threshold trigger has no matched, dated bilateral observation. | FAIL CLOSED — manual, probe, or source evidence required |
First-party provenance: Groq Cloud official pricing; verification date 2026-09-07. Missing or conflicting joins fail closed.
Method and limitations: formulas are deterministic; observed and assumed inputs are labeled; no missing provider, host, account, region, realm, alias, snapshot, revision, weight, artifact, control, tool, modality, workload, rate-period, timestamp, or result is transferred. Run this scenario →
Verified Model Pricing Intelligence•Audit date: 2026-09-08
GPT-OSS 20B on Groq API Pricing: Blazing Speed at Budget Rates
GPT-OSS 20B on Groq costs $0.07 per million input tokens and $0.30 per million output tokens ($0.1275/M blended at 3:1). Delivers lightning-fast inference on Groq LPUs for classification and code generation. Verified 2026-09-08.
LPU high-throughput token pricing and monthly spend matrix
Frozen scenario board. Formula / deterministic rule: monthly_spend = calls * ((in_tokens * 0.07 + out_tokens * 0.30) / 1M) Boundary: Owns Groq LPU token expenditure calculations for GPT-OSS 20B.
| Frozen scenario | Model, identity, provider, and evidence fields | Result | State |
|---|---|---|---|
| 50K low-latency conversational replies | model=gpt-oss-20b; slug=gpt-oss-20b; provider=Groq; scenario=50K low-latency conversational replies; workload; prompt tokens; completion tokens; monthly spend; speed advantage; cost verdict; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-08; measurement versus assumption=explicit | Unavailable — 50K low-latency conversational replies is a frozen fixture pending exact identity, configuration, and denominator joins. | UNTESTED — assumption cannot establish a verdict |
| 200K code generation tasks | model=gpt-oss-20b; slug=gpt-oss-20b; provider=Groq; scenario=200K code generation tasks; workload; prompt tokens; completion tokens; monthly spend; speed advantage; cost verdict; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-08; measurement versus assumption=explicit | Unavailable — 200K code generation tasks is a frozen fixture pending exact identity, configuration, and denominator joins. | UNTESTED — assumption cannot establish a verdict |
| 1M automated customer triage passes | model=gpt-oss-20b; slug=gpt-oss-20b; provider=Groq; scenario=1M automated customer triage passes; workload; prompt tokens; completion tokens; monthly spend; speed advantage; cost verdict; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-08; measurement versus assumption=explicit | Unavailable — 1M automated customer triage passes is a frozen fixture pending exact identity, configuration, and denominator joins. | UNTESTED — assumption cannot establish a verdict |
| high-concurrency request surge | model=gpt-oss-20b; slug=gpt-oss-20b; provider=Groq; scenario=high-concurrency request surge; workload; prompt tokens; completion tokens; monthly spend; speed advantage; cost verdict; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-08; measurement versus assumption=explicit | Unavailable — high-concurrency request surge is a frozen fixture pending exact identity, configuration, and denominator joins. | UNTESTED — assumption cannot establish a verdict |
| batch offline ingestion queue | model=gpt-oss-20b; slug=gpt-oss-20b; provider=Groq; scenario=batch offline ingestion queue; workload; prompt tokens; completion tokens; monthly spend; speed advantage; cost verdict; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-08; measurement versus assumption=explicit | Unavailable — batch offline ingestion queue is a frozen fixture pending exact identity, configuration, and denominator joins. | UNTESTED — assumption cannot establish a verdict |
| unresolved billing currency | model=gpt-oss-20b; slug=gpt-oss-20b; provider=Groq; scenario=unresolved billing currency; workload; prompt tokens; completion tokens; monthly spend; speed advantage; cost verdict; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-08; measurement versus assumption=explicit | Unavailable — unresolved billing currency has no matched, dated bilateral observation. | FAIL CLOSED — manual, probe, or source evidence required |
First-party provenance: Groq official API pricing; verification date 2026-09-08. Missing or conflicting joins fail closed.
Inference turnaround latency and streaming UX SLA audit
Frozen scenario board. Formula / deterministic rule: turnaround = ttft + (tokens_out / tps); Groq LPU ultra-high generation throughput Boundary: Owns turnaround time benchmarks and user experience responsiveness.
| Frozen scenario | Model, identity, provider, and evidence fields | Result | State |
|---|---|---|---|
| real-time conversational streaming (<80ms) | model=gpt-oss-20b; slug=gpt-oss-20b; provider=Groq; scenario=real-time conversational streaming (<80ms); task; response SLA; TTFT; generation speed; SLA compliance; responsiveness champion; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-08; measurement versus assumption=explicit | Unavailable — real-time conversational streaming (<80ms) is a frozen fixture pending exact identity, configuration, and denominator joins. | UNTESTED — assumption cannot establish a verdict |
| automated code autocomplete (<150ms) | model=gpt-oss-20b; slug=gpt-oss-20b; provider=Groq; scenario=automated code autocomplete (<150ms); task; response SLA; TTFT; generation speed; SLA compliance; responsiveness champion; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-08; measurement versus assumption=explicit | Unavailable — automated code autocomplete (<150ms) is a frozen fixture pending exact identity, configuration, and denominator joins. | UNTESTED — assumption cannot establish a verdict |
| instant document summary (<300ms) | model=gpt-oss-20b; slug=gpt-oss-20b; provider=Groq; scenario=instant document summary (<300ms); task; response SLA; TTFT; generation speed; SLA compliance; responsiveness champion; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-08; measurement versus assumption=explicit | Unavailable — instant document summary (<300ms) is a frozen fixture pending exact identity, configuration, and denominator joins. | UNTESTED — assumption cannot establish a verdict |
| long-form synthetic data generation | model=gpt-oss-20b; slug=gpt-oss-20b; provider=Groq; scenario=long-form synthetic data generation; task; response SLA; TTFT; generation speed; SLA compliance; responsiveness champion; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-08; measurement versus assumption=explicit | Unavailable — long-form synthetic data generation is a frozen fixture pending exact identity, configuration, and denominator joins. | UNTESTED — assumption cannot establish a verdict |
| network contention latency buffer | model=gpt-oss-20b; slug=gpt-oss-20b; provider=Groq; scenario=network contention latency buffer; task; response SLA; TTFT; generation speed; SLA compliance; responsiveness champion; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-08; measurement versus assumption=explicit | Unavailable — network contention latency buffer is a frozen fixture pending exact identity, configuration, and denominator joins. | UNTESTED — assumption cannot establish a verdict |
| unmeasured speed fixture | model=gpt-oss-20b; slug=gpt-oss-20b; provider=Groq; scenario=unmeasured speed fixture; task; response SLA; TTFT; generation speed; SLA compliance; responsiveness champion; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-08; measurement versus assumption=explicit | Unavailable — unmeasured speed fixture is a frozen fixture pending exact identity, configuration, and denominator joins. | UNTESTED — assumption cannot establish a verdict |
First-party provenance: Groq developer documentation; verification date 2026-09-08. Missing or conflicting joins fail closed.
Model tier escalation and multi-model routing model
Frozen scenario board. Formula / deterministic rule: blended_cost = 20b_volume * 20b_cost + 120b_volume * 120b_cost Boundary: Owns two-tier architectural routing between fast 20B triage and deep 120B generation.
| Frozen scenario | Model, identity, provider, and evidence fields | Result | State |
|---|---|---|---|
| 100% GPT-OSS 20B baseline ($0.07/$0.30) | model=gpt-oss-20b; slug=gpt-oss-20b; provider=Groq; scenario=100% GPT-OSS 20B baseline ($0.07/$0.30); escalation mix; total monthly queries; blended spend; cost savings vs pure 120B; accuracy retention; routing recommendation; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-08; measurement versus assumption=explicit | Unavailable — 100% GPT-OSS 20B baseline ($0.07/$0.30) is a frozen fixture pending exact identity, configuration, and denominator joins. | UNTESTED — assumption cannot establish a verdict |
| 90% 20B triage / 10% 120B escalation | model=gpt-oss-20b; slug=gpt-oss-20b; provider=Groq; scenario=90% 20B triage / 10% 120B escalation; escalation mix; total monthly queries; blended spend; cost savings vs pure 120B; accuracy retention; routing recommendation; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-08; measurement versus assumption=explicit | Unavailable — 90% 20B triage / 10% 120B escalation is a frozen fixture pending exact identity, configuration, and denominator joins. | UNTESTED — assumption cannot establish a verdict |
| 80% 20B triage / 20% 120B escalation | model=gpt-oss-20b; slug=gpt-oss-20b; provider=Groq; scenario=80% 20B triage / 20% 120B escalation; escalation mix; total monthly queries; blended spend; cost savings vs pure 120B; accuracy retention; routing recommendation; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-08; measurement versus assumption=explicit | Unavailable — 80% 20B triage / 20% 120B escalation is a frozen fixture pending exact identity, configuration, and denominator joins. | UNTESTED — assumption cannot establish a verdict |
| 50% 20B triage / 50% 120B escalation | model=gpt-oss-20b; slug=gpt-oss-20b; provider=Groq; scenario=50% 20B triage / 50% 120B escalation; escalation mix; total monthly queries; blended spend; cost savings vs pure 120B; accuracy retention; routing recommendation; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-08; measurement versus assumption=explicit | Unavailable — 50% 20B triage / 50% 120B escalation is a frozen fixture pending exact identity, configuration, and denominator joins. | UNTESTED — assumption cannot establish a verdict |
| 100% direct GPT-OSS 120B execution | model=gpt-oss-20b; slug=gpt-oss-20b; provider=Groq; scenario=100% direct GPT-OSS 120B execution; escalation mix; total monthly queries; blended spend; cost savings vs pure 120B; accuracy retention; routing recommendation; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-08; measurement versus assumption=explicit | Unavailable — 100% direct GPT-OSS 120B execution is a frozen fixture pending exact identity, configuration, and denominator joins. | UNTESTED — assumption cannot establish a verdict |
| unresolved confidence threshold trigger | model=gpt-oss-20b; slug=gpt-oss-20b; provider=Groq; scenario=unresolved confidence threshold trigger; escalation mix; total monthly queries; blended spend; cost savings vs pure 120B; accuracy retention; routing recommendation; prompt/config/input/output/result/cache/checkpoint/artifact hashes=required; evidence=2026-09-08; measurement versus assumption=explicit | Unavailable — unresolved confidence threshold trigger has no matched, dated bilateral observation. | FAIL CLOSED — manual, probe, or source evidence required |
First-party provenance: Groq official API pricing; verification date 2026-09-08. Missing or conflicting joins fail closed.
Exact-Model Pricing Evidence•Verified 2026-04-06; revalidation required before current claims.
GPT-OSS 20B on Groq pricing evidence
Source-backed dated rate shape: dated input=$0.075000/M; output=$0.300000/M. Groq-hosted GPT-OSS 20B unit economics only; 120B and Cerebras hosting, edge hardware, and cheapest-model verdicts retain their owners. Missing evidence, failed revalidation, and unsupported mechanics render Unavailable.
Module 1 of 3: Classification and autocomplete bill ladder
Novel contribution boundary: Owns dated Groq-hosted 20B arithmetic for fixed utility shapes; classification quality and latency are not asserted. Formula / deterministic rule: spend = requests × (inputTokens × inputCostPer1k + outputTokens × outputCostPer1k) / 1,000
| Scenario | Exact model, provider, dated rates, and fixed inputs | Result | State |
|---|---|---|---|
| 100K classification requests | model=gpt-oss-20b; provider=groq; registryRates=dated input=$0.075000/M; output=$0.300000/M; comparisonModel=gpt-oss-120b; fixedInputs=requests=100,000; inputTokens=300; outputTokens=80; FAIL CLOSED — no revalidated observation or provider guarantee; measurement versus assumption=explicit | Unavailable — Last verified 2026-04-06, before revalidation date 2026-09-14. | FAIL CLOSED — no revalidated observation or provider guarantee |
| 1M autocomplete requests | model=gpt-oss-20b; provider=groq; registryRates=dated input=$0.075000/M; output=$0.300000/M; comparisonModel=gpt-oss-120b; fixedInputs=requests=1,000,000; inputTokens=600; outputTokens=120; FAIL CLOSED — no revalidated observation or provider guarantee; measurement versus assumption=explicit | Unavailable — Last verified 2026-04-06, before revalidation date 2026-09-14. | FAIL CLOSED — no revalidated observation or provider guarantee |
| 10M compact extraction requests | model=gpt-oss-20b; provider=groq; registryRates=dated input=$0.075000/M; output=$0.300000/M; comparisonModel=gpt-oss-120b; fixedInputs=requests=10,000,000; inputTokens=1,000; outputTokens=200; FAIL CLOSED — no revalidated observation or provider guarantee; measurement versus assumption=explicit | Unavailable — Last verified 2026-04-06, before revalidation date 2026-09-14. | FAIL CLOSED — no revalidated observation or provider guarantee |
First-party provenance: https://console.groq.com/docs/models; registry provider groq; verified 2026-04-06; freshness gate=2026-09-14. Revalidate before any current-price claim.
Module 2 of 3: 20B→120B accepted-result premium
Novel contribution boundary: Owns dated cross-model cost arithmetic with explicit retry input; accepted-result quality and routing policy are not sourced. Formula / deterministic rule: requiredAcceptedRate = 120BAttemptCost / 20BAttemptCost; observed acceptance = Unavailable
| Scenario | Exact model, provider, dated rates, and fixed inputs | Result | State |
|---|---|---|---|
| Short classification; 10% retry share; 100 retry-input tokens | model=gpt-oss-20b; provider=groq; registryRates=dated input=$0.075000/M; output=$0.300000/M; comparisonModel=gpt-oss-120b; fixedInputs=requests=1; inputTokens=300; outputTokens=80; retryInputTokens=100; retryShare=10%; FAIL CLOSED — no revalidated observation or provider guarantee; measurement versus assumption=explicit | Unavailable — Last verified 2026-04-06, before revalidation date 2026-09-14. | FAIL CLOSED — no revalidated observation or provider guarantee |
| Medium extraction; 20% retry share; 250 retry-input tokens | model=gpt-oss-20b; provider=groq; registryRates=dated input=$0.075000/M; output=$0.300000/M; comparisonModel=gpt-oss-120b; fixedInputs=requests=1; inputTokens=600; outputTokens=120; retryInputTokens=250; retryShare=20%; FAIL CLOSED — no revalidated observation or provider guarantee; measurement versus assumption=explicit | Unavailable — Last verified 2026-04-06, before revalidation date 2026-09-14. | FAIL CLOSED — no revalidated observation or provider guarantee |
| Long autocomplete; 30% retry share; 500 retry-input tokens | model=gpt-oss-20b; provider=groq; registryRates=dated input=$0.075000/M; output=$0.300000/M; comparisonModel=gpt-oss-120b; fixedInputs=requests=1; inputTokens=1,000; outputTokens=200; retryInputTokens=500; retryShare=30%; FAIL CLOSED — no revalidated observation or provider guarantee; measurement versus assumption=explicit | Unavailable — Last verified 2026-04-06, before revalidation date 2026-09-14. | FAIL CLOSED — no revalidated observation or provider guarantee |
First-party provenance: https://console.groq.com/docs/models; registry provider groq; verified 2026-04-06; freshness gate=2026-09-14. Revalidate before any current-price claim.
Module 3 of 3: LPU versus local deployment input ledger
Novel contribution boundary: Owns exact-model registry evidence only; VRAM, utilization, energy, local hardware, and quota are not inferred. Formula / deterministic rule: localTCO = hardware + energy + operations; hardware, utilization, and quota inputs = Unavailable
| Scenario | Exact model, provider, dated rates, and fixed inputs | Result | State |
|---|---|---|---|
| VRAM or local hardware requirement | model=gpt-oss-20b; provider=groq; registryRates=dated input=$0.075000/M; output=$0.300000/M; comparisonModel=gpt-oss-120b; fixedInputs=evidenceField=hardware; FAIL CLOSED — no revalidated observation or provider guarantee; measurement versus assumption=explicit | Unavailable — Last verified 2026-04-06, before revalidation date 2026-09-14. | FAIL CLOSED — no revalidated observation or provider guarantee |
| LPU quota or rate-limit field | model=gpt-oss-20b; provider=groq; registryRates=dated input=$0.075000/M; output=$0.300000/M; comparisonModel=gpt-oss-120b; fixedInputs=evidenceField=quota; FAIL CLOSED — no revalidated observation or provider guarantee; measurement versus assumption=explicit | Unavailable — Last verified 2026-04-06, before revalidation date 2026-09-14. | FAIL CLOSED — no revalidated observation or provider guarantee |
| Utilization or energy input | model=gpt-oss-20b; provider=groq; registryRates=dated input=$0.075000/M; output=$0.300000/M; comparisonModel=gpt-oss-120b; fixedInputs=evidenceField=throughput; FAIL CLOSED — no revalidated observation or provider guarantee; measurement versus assumption=explicit | Unavailable — Last verified 2026-04-06, before revalidation date 2026-09-14. | FAIL CLOSED — no revalidated observation or provider guarantee |
First-party provenance: https://console.groq.com/docs/models; registry provider groq; verified 2026-04-06; freshness gate=2026-09-14. Revalidate before any current-price claim.
How fast is GPT-OSS 20B?
Tokens / sec
1120
TTFT
140 ms
Rank
#3 of 31
$ / M ÷ t/s
$0.0001
Measured with 5 runs on a fixed prompt — see the full methodology.
How much does GPT-OSS 20B cost at scale?
| Tokens / month | Est. cost (blended 3:1) |
|---|---|
| 100,000 | $0.01 |
| 1,000,000 | $0.13 |
| 10,000,000 | $1.31 |
| 100,000,000 | $13.12 |
How does GPT-OSS 20B compare with other models?
GPT-OSS 120B — $0.26/MLlama 4 Maverick — $0.30/MQwen 3.8 30B — $1.20/MQwen 3.6 27B — $1.20/MMuse Spark 1.3 Contributor — $0.13/MGPT-5 Nano — $0.14/MMinistral 8B — $0.15/M
What is GPT-OSS 20B best for?
What should you explore next for GPT-OSS 20B?
What are common questions about GPT-OSS 20B?
Is GPT-OSS 20B cheaper than Muse Spark 1.3 Contributor?
GPT-OSS 20B costs $0.13/M blended tokens, Muse Spark 1.3 Contributor costs $0.13/M — Muse Spark 1.3 Contributor is cheaper.
How much does 1 million tokens cost with GPT-OSS 20B?
At a 3:1 input:output ratio, 1 million blended tokens costs approximately $0.13. Pure input costs $0.07/M; pure output costs $0.30/M.
What does GPT-OSS 20B cost at high volume?
At 100 million blended tokens a month, GPT-OSS 20B costs approximately $13.12. See the cost-at-scale table below for other volumes.