Groq Introduces Llama 4 Maverick and Qwen3 Models Amid Removal of Multiple High-Cost Plans and End of 50% Batch Processing Discount
What happened
Between the Apr 1, 2025 and Jul 4, 2025 captures of Groq's pricing page, Pulse recorded 19 changes: Llama 3.2 11B Vision 8k (Preview) model removed (was $0.18 input/$0.18 output); DeepSeek R1 Distill Qwen 32B model removed (was $0.69 input/$0.69 output); Qwen 2.5 32B Instruct model removed (was $0.79 input/$0.79 output); Llama 4 Maverick (17Bx128E) model added with 562 tokens/sec speed, $0.20 input/$0.60 output per M tokens; Llama 3.2 3B (Preview) model removed (was $0.06 input/$0.06 output); Qwen3 32B model added with 662 tokens/sec speed, $0.29 input/$0.59 output per M tokens; Qwen 2.5 Coder 32B Instruct model removed (was $0.79 input/$0.79 output); Llama 3.3 70B SpecDec model removed (was $0.59 input/$0.99 output); Whisper Large v3 Turbo speed factor increased from 216x to 228x; Llama 3.2 90B Vision 8k (Preview) model removed (was $0.90 input/$0.90 output); Llama 3.1 8B Instant speed increased from 750 to 840 tokens/sec; Llama 3.2 1B (Preview) model removed (was $0.04 input/$0.04 output); Llama 4 Scout (17Bx16E) model added with 594 tokens/sec speed, $0.11 input/$0.34 output per M tokens; Llama Guard 4 12B model added with 325 tokens/sec speed, $0.20 input/$0.20 output per M tokens; Whisper V3 Large speed factor increased from 189x to 217x; Llama 3 8B speed increased from 1250 to 1345 tokens/sec; DeepSeek R1 Distill Llama 70B speed increased from 275 to 400 tokens/sec; 50% batch processing discount promotion ended (was doubled from 25% through end of April 2025); Llama 3.3 70B Versatile speed increased from 275 to 394 tokens/sec. Pulse files it as a pricing change (plan removed, new plan, limit increase, discount removed) with a downward direction. Pulse's latest capture (Sep 20, 2026) lists GPT OSS 20B at $0.08 per 1M input tokens + $0.3 per 1M output tokens, Whisper Large v3 (ASR) at $0.11 per hour transcribed, Whisper Large V3 Turbo (ASR) at $0.04 per hour transcribed, Canopy Labs Orpheus Arabic Saudi (TTS) at $40 per 1M characters and Canopy Labs Orpheus V1 English (TTS) at $22 per 1M characters.
Groq's previous recorded change was "Groq Introduces New High-Performance Models and 50% Discount on Batch API for Paid GroqCloud Customers" (Q1 2025). Groq has been tracked since 2024-11-17.
In the same window Anthropic also changed their AI & Machine Learning pricing pages. Those moves are listed further down, and the Pulse Agent can compare them side by side.
- — · PlanPlan removed
Llama 3.2 11B Vision 8k (Preview) model removed (was $0.18 input/$0.18 output)
- — · PlanPlan removed
DeepSeek R1 Distill Qwen 32B model removed (was $0.69 input/$0.69 output)
- — · PlanPlan removed
Qwen 2.5 32B Instruct model removed (was $0.79 input/$0.79 output)
- — · PlanNew plan
Llama 4 Maverick (17Bx128E) model added with 562 tokens/sec speed, $0.20 input/$0.60 output per M tokens
- — · PlanPlan removed
Llama 3.2 3B (Preview) model removed (was $0.06 input/$0.06 output)
- — · PlanAddedNo Qwen3 32B modelQwen3 32B model
- — · PlanPlan removed
Qwen 2.5 Coder 32B Instruct model removed (was $0.79 input/$0.79 output)
- Llama 3.3 70B · PlanPlan removed
Llama 3.3 70B SpecDec model removed (was $0.59 input/$0.99 output)
- — · LimitChanged216x228x
- — · PlanPlan removed
Llama 3.2 90B Vision 8k (Preview) model removed (was $0.90 input/$0.90 output)
- Llama 3.1 8B · Limit+12%750840 tokens/sec
- — · PlanPlan removed
Llama 3.2 1B (Preview) model removed (was $0.04 input/$0.04 output)
- — · PlanNew plan
Llama 4 Scout (17Bx16E) model added with 594 tokens/sec speed, $0.11 input/$0.34 output per M tokens
- — · PlanAdded
BeforeNo Llama Guard 4 12B model
AfterLlama Guard 4 12B model
- — · LimitChanged189x217x
- — · Limit+8%12501345 tokens/sec
- — · Limit+45%275400 tokens/sec
- — · DiscountRemovedDiscountNo discount
- Llama 3.3 70B · Limit+43%275394 tokens/sec
| Plan | Field | Before | After | Change |
|---|---|---|---|---|
| — | Plan | Llama 3.2 11B Vision 8k (Preview) model removed (was $0.18 input/$0.18 output) | Plan removed | |
| — | Plan | DeepSeek R1 Distill Qwen 32B model removed (was $0.69 input/$0.69 output) | Plan removed | |
| — | Plan | Qwen 2.5 32B Instruct model removed (was $0.79 input/$0.79 output) | Plan removed | |
| — | Plan | Llama 4 Maverick (17Bx128E) model added with 562 tokens/sec speed, $0.20 input/$0.60 output per M tokens | New plan | |
| — | Plan | Llama 3.2 3B (Preview) model removed (was $0.06 input/$0.06 output) | Plan removed | |
| — | Plan | No Qwen3 32B model | Qwen3 32B model | Added |
| — | Plan | Qwen 2.5 Coder 32B Instruct model removed (was $0.79 input/$0.79 output) | Plan removed | |
| Llama 3.3 70B | Plan | Llama 3.3 70B SpecDec model removed (was $0.59 input/$0.99 output) | Plan removed | |
| — | Limit | 216x | 228x | Changed |
| — | Plan | Llama 3.2 90B Vision 8k (Preview) model removed (was $0.90 input/$0.90 output) | Plan removed | |
| Llama 3.1 8B | Limit | 750 | 840 tokens/sec | +12% |
| — | Plan | Llama 3.2 1B (Preview) model removed (was $0.04 input/$0.04 output) | Plan removed | |
| — | Plan | Llama 4 Scout (17Bx16E) model added with 594 tokens/sec speed, $0.11 input/$0.34 output per M tokens | New plan | |
| — | Plan | No Llama Guard 4 12B model | Llama Guard 4 12B model | Added |
| — | Limit | 189x | 217x | Changed |
| — | Limit | 1250 | 1345 tokens/sec | +8% |
| — | Limit | 275 | 400 tokens/sec | +45% |
| — | Discount | Discount | No discount | Removed |
| Llama 3.3 70B | Limit | 275 | 394 tokens/sec | +43% |
Values as written on Groq's public pricing page in the two captures Pulse compared; where the page states no number, the cells give the wording before and after.
Evidence
The two captures Pulse comparedFull-page captures of Groq's pricing page (groq.com) from Apr 1, 2025 and Jul 4, 2025; the change is marked on the page, where Qwen3 32B model now appears.
Sign in to see the evidence
The two full-page captures with this change marked up are free with an account.
Groq, the last 90 days
Full history- Jan 10, 2026decreaseGroq Reduces GPT OSS 20B Input and Output Token Prices by 25% and 40%, Enhancing Cost Efficiency for Users
- Week 1, 2026packagingGroq swaps PlayAI for Orpheus TTS, cuts text-to-speech price 56%
- Q3 2025packagingGroq Removes Multiple Models While Adding High-Performance GPT OSS 20B and 120B with New Prompt Caching Feature
- Q2 2025improvedGroq Introduces Llama 4 Maverick and Qwen3 Models Amid Removal of Multiple High-Cost Plans and End of 50% Batch Processing Discount · this change
- Q1 2025packagingGroq Introduces New High-Performance Models and 50% Discount on Batch API for Paid GroqCloud Customers
Across AI & Machine Learning in the same window
All AI & Machine Learning changes- AnthropicJan 10Anthropic API Introduces Tiered Pricing and Prompt Caching for Sonnet 4.5, Restructuring Costs Based on Token Length
- AnthropicJan 10Anthropic API Introduces Code Execution Tool with 50 Free Hours Daily, Enhancing Service Offerings and Flexibility for Users
- AnthropicJan 10Anthropic API Cuts Flagship Opus Model Prices by 67% in Major Update to Version 4.5, Launching New Pricing Structure
- AnthropicJan 10Anthropic Max Plan Now Includes Memory Across Conversations, Enhancing User Experience and Functionality
- AnthropicJan 10Anthropic Introduces Max Plan at $100+/Month, Offering 5x-20x More Usage Than Pro Plan
Questions about this change
Did Groq raise prices in July 2025?
No list-price increase was recorded between Apr 1, 2025 and Jul 4, 2025. The change Pulse detected was a pricing change: plan removed, new plan, limit increase, discount removed.
What changed on Groq's pricing page between Apr 1, 2025 and Jul 4, 2025?
Llama 3.2 11B Vision 8k (Preview) model removed (was $0.18 input/$0.18 output); DeepSeek R1 Distill Qwen 32B model removed (was $0.69 input/$0.69 output); Qwen 2.5 32B Instruct model removed (was $0.79 input/$0.79 output); Llama 4 Maverick (17Bx128E) model added with 562 tokens/sec speed, $0.20 input/$0.60 output per M tokens; Llama 3.2 3B (Preview) model removed (was $0.06 input/$0.06 output); plan: No Qwen3 32B model → Qwen3 32B model, added; Qwen 2.5 Coder 32B Instruct model removed (was $0.79 input/$0.79 output); Llama 3.3 70B SpecDec model removed (was $0.59 input/$0.99 output); limit: 216x → 228x, changed; Llama 3.2 90B Vision 8k (Preview) model removed (was $0.90 input/$0.90 output); Llama 3.1 8B limit: 750 → 840 tokens/sec (+12%); Llama 3.2 1B (Preview) model removed (was $0.04 input/$0.04 output); Llama 4 Scout (17Bx16E) model added with 594 tokens/sec speed, $0.11 input/$0.34 output per M tokens; plan: No Llama Guard 4 12B model → Llama Guard 4 12B model, added; limit: 189x → 217x, changed; limit: 1250 → 1345 tokens/sec (+8%); limit: 275 → 400 tokens/sec (+45%); discount: Discount → No discount, removed; Llama 3.3 70B limit: 275 → 394 tokens/sec (+43%). Pulse compared the captures it took on Apr 1, 2025 and Jul 4, 2025; the before/after table above lists each item as written on the page.
How often does Groq change its pricing?
Pulse recorded 1 other pricing-page change for Groq in the 90 days before this one; the previous change was "Groq Introduces New High-Performance Models and 50% Discount on Batch API for Paid GroqCloud Customers" in Q1 2025; Groq has been tracked since 2024-11-17.