API Rate Limits Compared: Every Major LLM Provider (October 2026)
API rate limits for every major LLM provider — October 17, 2026. Side-by-side tables for OpenAI, Anthropic, Google, Groq, xAI, DeepSeek, Mistral, Cerebras, SambaNova, and more.
Eleventh edition. The tenth (September 10, 2026) covered 17 providers. This update reflects the five-week window since September 10. Google’s current flagship is now Gemini 4 Argon (GA, October 13, 2026), replacing Gemini 3.8 Flash at the top of the lineup. Anthropic’s current lineup is Fable 5.1 / Opus 5.5 / Sonnet 5.5 / Haiku 5.5 — all launched since the last edition. OpenAI’s current flagship is GPT-6 Astra; GPT-6 Luna replaces the now-deprecated GPT-4.1 family as the budget/high-throughput tier. xAI’s current flagship is Grok 4.7. DeepSeek’s current flagship is DeepSeek-V4.1-Flash. Mistral Large 4 is in public preview. Qwen3.8-Max weights have been released openly. Rate limit tables for most new models remain unpublished — predecessor figures are carried forward as provisional floors throughout.
Table of Contents
- What changed since September 10
- Free Tier — RPM & RPD
- Free Tier — TPM & TPD
- Entry Paid Tier — RPM & RPD
- Entry Paid Tier — TPM
- Scaled Tier — RPM & TPM
- Cerebras & SambaNova
- Audio Models — ASH & ASD
- Cloud Aggregators — Azure AI & AWS Bedrock
- More Providers — Perplexity, Alibaba, Moonshot
- Provider Notes
- Tips for managing rate limits
- Further Reading
Last updated: October 17, 2026.
What changed since September 10
Google’s current flagship is now Gemini 4 Argon. GA October 13, 2026. Model ID gemini-4-argon; 1M unified context with long decode up to 1M output tokens (65,536 by default); ranked #1 on the Vals Index; built for complex software engineering, deep mathematical and scientific problem solving, and autonomous multi-step agents. Available via the Gemini API and AI Studio, to AI Ultra subscribers in Antigravity, and to AI Ultra and Pro subscribers in the Gemini app and AI Mode in Search. Paid tier only. Introductory pricing $2/$10 per M input/output through November 30, 2026, doubling to $4/$20 on January 1, 2027 (a larger-prompt tier also exists at $4/$20, but its exact token threshold is not confirmed — do not quote a specific cutoff). Rate limit figures for Gemini 4 Argon are not yet published in public documentation. Gemini 3.8 Flash (GA, September 2, 2026) is now the most intelligent Flash model rather than the overall flagship; Gemini 3.7 Flash remains the default for Antigravity.
Gemini 3.8 Flash is now GA. Model ID gemini-3.8-flash; 1M context, 64K max output, tunable thinking levels low/medium/high; introductory pricing $0.75/$3.75 per M through December 31, 2026, rising to $1.50/$7.50 on January 1, 2027. Gemini 3.8 Flash Cyber (cybersecurity domain) launched alongside it. Rate limit figures for Gemini 3.8 Flash are not yet published in public documentation.
OpenAI’s current flagship is GPT-6 Astra. Released September 2026. GPT-6.1 Sol (September 29; $2/$10 per M, cached $0.10) nearly matches Astra on agentic coding, computer use, and professional work at approximately one-fifth of Astra’s token prices. GPT-6 Luna (September 22; $0.10/$0.50 per M, cached $0.01) is OpenAI’s cheapest current model and the stated replacement for the retiring GPT-4.1 family. Rate limit tier tables for GPT-6 Astra, GPT-6.1 Sol, and GPT-6 Luna are not yet published in public documentation — GPT-5.5 figures are carried forward as a provisional floor throughout.
GPT-5.6 Sol repriced. GPT-5.6 Sol dropped from $5/$30 to $4/$20 per M after the GPT-6 launch. Any cost model carrying the old $5/$30 figure needs revision. GPT-5.5 ($5/$30) is unchanged.
GPT-4.1 family deprecated; GPT-5.4 Nano deprecated. GPT-4.1 and GPT-4.1 Mini were deprecated October 14, 2026 and shut down January 14, 2027. GPT-4.1 Nano was deprecated April 22, 2026 and shuts down October 23, 2026 — within days of this edition. GPT-5.4 Nano was deprecated October 1, 2026 and shuts down April 1, 2027. OpenAI names GPT-6 Luna as the replacement for all of these. Any pipeline still referencing these models should migrate immediately. Posts from before October 2026 that named GPT-4.1 Nano as the budget tier were accurate at their publication date and should not be retro-edited.
Anthropic’s current lineup is Fable 5.1 / Opus 5.5 / Sonnet 5.5 / Haiku 5.5. All 1M context, 128K max output, June 2026 reliable knowledge cutoff. Claude Opus 5.5 (September 22; $4/$20 per M, cache reads $0.20) is Anthropic’s recommended starting point for most workloads. Claude Fable 5.1 (September 1; $10/$50 per M, cache reads $0.25) is the top-capability tier for demanding reasoning and long-horizon agentic work. Claude Sonnet 5.5 (September 28; $2/$10 per M, cache reads $0.10) is the current balanced tier. Claude Haiku 5.5 (October 7) is the current fast/budget tier with tiered pricing: $0.10/$0.50 per M for prompts up to 100,000 tokens, $0.50/$2.50 above that. Rate limit figures for all four new models are not yet published in public documentation — predecessor figures are carried forward as provisional floors.
xAI’s current flagship is Grok 4.7. Released September 21, 2026. Model ID grok-4.7; 500K context; $2/$6 per M below 200K prompt tokens, $4/$12 above. Grok 4.6 is the previous flagship, still available. Numerical RPM/TPM remain console-only — not published in public documentation.
DeepSeek’s current flagship is DeepSeek-V4.1-Flash. Released September 10, 2026. Model ID deepseek-flash; 552B MoE with 8B active for input and 16B for output; native multimodal visual understanding; 1M context, 384K max output. The older deepseek-v4-flash and deepseek-v4-flash-vision-exp names are retired and now route to V4.1-Flash. Pricing varies by time of day: $0.30/$1.20 per M at peak, $0.15/$0.60 at off-peak. DeepSeek V4-Pro traffic reportedly reroutes to V4.1-Flash at V4.1-Flash rates as of September 14, but the live pricing page still lists V4-Pro separately — treat V4-Pro pricing as uncertain and check both sources.
Mistral Large 4 is in public preview. Released October 6, 2026. Open-weight multimodal; 1.05T total / 52B active parameters plus a 1.6B vision encoder; 1M context; strong on agentic coding. Public preview, not GA. Per-token pricing not confirmed — do not quote one.
Qwen3.8-Max weights released openly. Alibaba released the flagship’s weights as Qwen3.8-2.4T-A95B on Hugging Face and ModelScope — the first Qwen-Max-class model to ship openly. Prior guidance calling it closed-weight is now obsolete. Published per-token pricing remains absent — write “not published” rather than guessing.
Cohere Command A+ licence unconfirmed. The previous edition stated Apache 2.0 for Command A+. That licence was not confirmed in the current sweep. Do not state a licence for Command A+ until it is verified on Cohere’s model page.
Tokenizer note corrected. The ~30%-more-tokens tokenizer arrived with the Claude 4.7 generation, not with Sonnet 5 as stated in earlier editions. Claude 4.7 and later use it; Sonnet 4.6 and earlier use the previous one.
Introductory pricing expiry calendar — updated:
- Gemini 4 Argon: introductory $2/$10 per M through November 30, 2026 — standard rates double January 1, 2027.
- Gemini 3.8 Flash, 3.7 Flash, 3.6 Flash: introductory $0.75/$3.75 per M through December 31, 2026 — standard $1.50/$7.50 from January 1, 2027. Cost models projecting past that date must use the standard rate.
Still pending formal documentation:
- OpenAI: GPT-6 Astra, GPT-6.1 Sol, and GPT-6 Luna rate limit tier tables not yet published. GPT-5.5 figures carried forward as a provisional floor.
- Anthropic: Claude Opus 5.5, Sonnet 5.5, Haiku 5.5, and Fable 5.1 rate limit figures not yet published. Predecessor figures carried forward as provisional floors.
- Google: Gemini 4 Argon rate limit figures not yet published. Gemini 3.8 Flash, 3.7 Flash, 3.6 Flash, 3.5 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber rate limit figures also unpublished. Gemini 3.5 Pro remains in limited preview.
- xAI: Grok 4.7 and Grok 4.6 numerical rate limit figures console-only, not in public documentation.
- DeepSeek: Fully dynamic concurrency limits; no published RPM/TPM for any model.
- Alibaba: Qwen3.8-Max rate limit figures not yet published. Qwen3.7-Max limits carried forward as a provisional floor.
- Moonshot: Kimi K3 rate limit figures not yet confirmed in public documentation. K2.5 figures carried forward.
- Mistral: Numerical limits require login at
admin.mistral.ai. Large 4 pricing unconfirmed.
Deprecation calendar:
- GPT-4.1 Nano shuts down October 23, 2026 — days away. Migrate to GPT-6 Luna now.
- GPT-4.1 and GPT-4.1 Mini deprecated October 14, 2026; shut down January 14, 2027.
- GPT-5.4 Nano deprecated October 1, 2026; shuts down April 1, 2027.
- OpenAI Assistants API deprecated August 2026 — passed. Migrate to the Responses API.
- DeepSeek V3.2 aliases deprecated July 24, 2026 — passed. Those endpoints are gone.
- Gemini 4 Argon introductory pricing expires November 30, 2026 — rates double January 1, 2027.
- Gemini 3.8 Flash / 3.7 Flash / 3.6 Flash introductory pricing expires December 31, 2026 — standard $1.50/$7.50 from January 1, 2027.
What are rate limits?
Rate limits cap how much you can use an API within a given time window. Providers enforce them to manage capacity and ensure fair access. Exceeding a limit returns a 429 Too Many Requests error.
Most providers use the token bucket algorithm: capacity refills continuously up to your maximum, rather than resetting at fixed intervals. A 60 RPM limit means roughly 1 request per second with steady refill, not 60 requests then a hard stop.
The metrics
- RPM (Requests per minute): API calls per minute, regardless of size. A 10-token request and a 100K-token request both count as one.
- RPD (Requests per day): Daily cap on total API calls. Some providers use this instead of (or alongside) RPM, especially on free tiers.
- TPM (Tokens per minute): Total tokens (input + output) processed per minute. Usually the binding constraint in production.
- TPD (Tokens per day): Daily token cap. More common on free tiers.
- ITPM/OTPM (Input/Output tokens per minute): Anthropic separates input and output token limits, giving finer control. Cached input tokens don’t count toward ITPM on current Claude 4.x and 5.x models.
- ASH/ASD (Audio seconds per hour/day): For speech models like Whisper.
RPM limits set your concurrency ceiling. TPM limits set your throughput. For batch workloads, RPD and TPD matter more. For real-time apps, RPM and TPM hit first.
Free Tier — Requests per Minute (RPM) & Requests per Day (RPD)
| Provider / Model | RPM | RPD |
|---|---|---|
| Google — Gemini 3.1 Flash-Lite | 15 | 1,000 |
| Groq — Llama 4 Scout | 30 | 1,000 |
| Cerebras — Llama 4 Scout | 30 | — |
| SambaNova — open models | 10–30 | — |
| Fireworks AI — OS models | 10 | — |
| Cohere — Command R+* | 20 | ~33† |
*Cohere’s free tier figures reference Command R+. Command A+ free tier limits not yet published. †Cohere trial keys get 1,000 calls/month across all endpoints, roughly ~33/day.
Note: Gemini 4 Argon, 3.8 Flash, 3.7 Flash, 3.6 Flash, 3.5 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber free tier rate limit figures are not yet published in public documentation. Gemini 3.1 Flash-Lite figures shown above are the last confirmed published numbers. OpenAI, Anthropic, xAI, DeepSeek, and Mistral have no free API tier for current flagship models. OpenAI’s free tier does not support any GPT-5.x or GPT-6.x model via the API. Anthropic’s lowest entry is $5 (Tier 1). xAI and Mistral gate numerical limits behind their respective consoles.
Migration note: Gemini 2.0 Flash and Flash-Lite shut down June 1, 2026. DeepSeek V3.2 aliases deprecated July 24, 2026. GPT-4.1 Nano shuts down October 23, 2026 — within days of this edition.
Free Tier — Tokens per Minute (TPM) & Tokens per Day (TPD)
| Provider / Model | TPM | TPD |
|---|---|---|
| Google — Gemini 3.1 Flash-Lite | 250,000 | — |
| Groq — Llama 4 Scout | 30,000 | 500,000 |
| Groq — other open models | 6,000–30,000 | 500,000 |
| Cerebras — Llama 4 Scout | 60,000 | 1,000,000 |
| SambaNova (free tier) | — | — |
Standout: Google leads with 250K TPM on the free tier for Gemini 3.1 Flash-Lite, but the 10–15 RPM ceiling means it’s best suited to fewer, larger requests. Gemini 4 Argon, 3.8 Flash, 3.7 Flash, 3.6 Flash, 3.5 Flash, and 3.5 Flash-Lite free tier TPM figures are not yet published. Cerebras offers 60K TPM with 1M TPD and approximately 2,100 tokens/second inference speed on Llama 4 Scout — the fastest free option by throughput. SambaNova publishes RPM but not TPM/TPD specifics.
Entry Paid Tier — Requests per Minute (RPM) & Requests per Day (RPD)
The tier most developers start with. OpenAI Tier 1 ($5), Anthropic Tier 1 ($5), Google Tier 1 (pay-as-you-go), and others at their lowest paid level.
| Provider / Model | RPM | RPD |
|---|---|---|
| OpenAI — GPT-6 Astra* | 500* | — |
| OpenAI — GPT-6 Luna* | 500* | — |
| Anthropic — Fable 5.1† | 50† | — |
| Anthropic — Opus 5.5† | 50† | — |
| Anthropic — Sonnet 5.5† | 50† | — |
| Anthropic — Haiku 5.5† | 50† | — |
| Google — Gemini 3.1 Pro | 150 | 1,000 |
| Google — Gemini 3.1 Flash-Lite | 300 | 1,500 |
| xAI — Grok 4.7 | Console‡ | — |
| DeepSeek — V4.1-Flash | Dynamic§ | — |
| Fireworks AI — OS models | ≤6,000 | — |
| Cohere — Command A+** | 500** | — |
*GPT-6 Astra, GPT-6.1 Sol, and GPT-6 Luna rate limit tier tables not yet published. 500 RPM carried forward from GPT-5.5 as a provisional floor. †Fable 5.1, Opus 5.5, Sonnet 5.5, and Haiku 5.5 rate limit tier tables not yet published. 50 RPM carried forward from their respective predecessors as a provisional floor. ‡xAI publishes tier thresholds ($0/$50/$250/$1K/$5K) but numerical RPM/TPM are only visible in the xAI Console after login. §DeepSeek uses fully dynamic concurrency limits based on server load. No fixed RPM/TPM published. **Command A+ entry paid tier limits not yet published. 500 RPM is the Command R+ figure, carried forward as a reference.
Gemini 4 Argon and 3.8 Flash note: Neither model has published rate limit figures in public documentation. Gemini 3.7 Flash (previous flagship), 3.6 Flash, and 3.5 Flash also have no published rate limit figures. Gemini 3.1 Pro and 3.1 Flash-Lite remain the previous generation with the last confirmed published figures. Check AI Studio (aistudio.google.com/rate-limit) for your project’s actual limits.
Mistral note: Mistral moved to console-only limits. Actual numbers require login at admin.mistral.ai/plateforme/limits. Mistral is excluded from RPM columns where no public figure exists.
Standout: Fireworks can spike to 6,000 RPM but it’s a dynamic ceiling, not guaranteed (soft limit starts at approximately 1 RPS and doubles hourly). Anthropic’s 50 RPM at Tier 1 is the lowest here, but jumps 20x to 1,000 RPM at Tier 2 ($40). The ~30%-more-tokens tokenizer now applies to all Claude 4.7-generation models and later — TPM headroom shrinks faster for any model in that range than it did on Sonnet 4.6 or earlier at the same RPM.
Entry Paid Tier — Tokens per Minute (TPM)
| Provider / Model | TPM |
|---|---|
| OpenAI — GPT-6 Astra* | 500K* |
| OpenAI — GPT-6 Luna* | 500K* |
| Anthropic — Fable 5.1† | 30K in / 8K out† |
| Anthropic — Opus 5.5† | 30K in / 8K out† |
| Anthropic — Sonnet 5.5† | 30K in / 8K out† |
| Anthropic — Haiku 5.5† | 50K in / 10K out† |
| Google — Gemini 3.1 Pro | 1M |
| Google — Gemini 3.1 Flash-Lite | 2M |
| DeepSeek — V4.1-Flash | Dynamic |
*GPT-6 Astra and GPT-6 Luna TPM figures not yet published. 500K carried forward from GPT-5.5 as a provisional floor. †Fable 5.1, Opus 5.5, Sonnet 5.5, and Haiku 5.5 TPM figures not yet published. Predecessor figures carried forward as provisional floors.
Standout: Google’s 3.1 Flash-Lite leads confirmed figures at 2M TPM. Anthropic’s current models all carry the ~30%-more-tokens tokenizer (Claude 4.7 generation and later), meaning identical prompts now consume more tokens against ITPM limits than they did on Sonnet 4.6 or earlier. Cached input tokens don’t count toward ITPM on Claude 4.x and 5.x models — with an 80% cache hit rate, effective throughput remains substantially higher than the raw ITPM figure.
Scaled Tier — RPM & TPM at Higher Spend
For teams past entry-level. OpenAI Tier 3 ($100+), Anthropic Tier 3 ($200+), Google Tier 2 ($250+).
| Provider (Tier) / Model | RPM | TPM |
|---|---|---|
| OpenAI (Tier 3) — GPT-6 Astra* | 5,000* | 2M* |
| OpenAI (Tier 3) — GPT-6 Luna* | 5,000* | 4M* |
| Anthropic (Tier 3) — Fable 5.1† | 2,000† | 800K in / 160K out† |
| Anthropic (Tier 3) — Opus 5.5† | 2,000† | 800K in / 160K out† |
| Anthropic (Tier 3) — Sonnet 5.5† | 2,000† | 800K in / 160K out† |
| Anthropic (Tier 3) — Haiku 5.5† | 2,000† | 1M in / 200K out† |
| Google (Tier 2) — Gemini 3.1 Pro | 1,000 | 2M |
| Google (Tier 2) — Gemini 3.1 Flash-Lite | 2,000 | 4M |
*GPT-6 Astra and GPT-6 Luna Tier 3 figures not yet published. GPT-5.5 figures carried forward as a provisional floor. †Fable 5.1, Opus 5.5, Sonnet 5.5, and Haiku 5.5 Tier 3 figures not yet published. Predecessor figures carried forward as provisional floors.
At the highest standard tiers (OpenAI Tier 5 at $1,000+, Anthropic Tier 4 at $400+):
| Provider (Tier) / Model | RPM | TPM |
|---|---|---|
| OpenAI (Tier 5) — GPT-6 Astra* | 15,000* | 40M* |
| OpenAI (Tier 5) — GPT-6 Luna* | 30,000* | 180M* |
| Anthropic (Tier 4) — Fable 5.1† | 4,000† | 2M in / 400K out† |
| Anthropic (Tier 4) — Opus 5.5† | 4,000† | 2M in / 400K out† |
| Anthropic (Tier 4) — Sonnet 5.5† | 4,000† | 2M in / 400K out† |
| Anthropic (Tier 4) — Haiku 5.5† | 4,000† | 4M in / 800K out† |
*GPT-6 Astra and GPT-6 Luna Tier 5 figures not yet published. GPT-5.5 figures carried forward as a provisional floor. GPT-6 Luna’s provisionally carried figure at Tier 5 inherits GPT-4.1 Nano’s 180M TPM — treat with extra caution until official figures publish. †Fable 5.1, Opus 5.5, Sonnet 5.5, and Haiku 5.5 Tier 4 figures not yet published. Predecessor figures carried forward as provisional floors. Confirm current limits in the Anthropic console.
Standout: OpenAI’s Tier 5 numbers remain the highest of any provider with published or provisionally carried figures. GPT-6 Luna at a provisionally carried 180M TPM and 30K RPM is built for high-volume classification and routing — but the GPT-4.1 Nano figures those numbers are inherited from are from a now-deprecated model, so treat them as a rough floor only. Anthropic’s Tier 4 caps at 4K RPM, but cached tokens don’t count toward ITPM — effective throughput can be substantially higher with good cache hit rates.
Model pool notes:
- Fable 5.1 pool assignment is not yet formally documented. Check your Anthropic dashboard.
- Opus 5.5 pool assignment is not yet formally documented. Opus 4.8, 4.7, and 4.6 shared one rate limit bucket — check your dashboard for how the 5.x Opus line is bucketed.
- Sonnet 5.5 and Sonnet 5 likely share a pool — not formally confirmed; check your dashboard.
- Sending traffic to multiple model versions within the same pool draws from one shared bucket, not separate ones.
Gemini 4 Argon and 3.8 Flash scaled tier note: Neither model has published rate limit figures. Gemini 3.7 Flash, 3.6 Flash, and 3.5 Flash are also without published figures. The 3.1 Flash-Lite and 3.1 Pro figures shown above remain the last confirmed published numbers for Google’s API.
Cerebras & SambaNova
Both specialize in custom silicon for inference speed.
Cerebras
Hardware: Wafer-Scale Engine (WSE-3). Fastest published inference: approximately 2,100 tokens/second on Llama 4 Scout.
| Tier | RPM | TPM | TPD |
|---|---|---|---|
| Free | 30 | 60,000 | 1,000,000 |
| Paid | Higher (contact sales) | Higher | Higher |
Hosts Llama 4 Scout and other open models. The speed advantage is real: tasks that take 30 seconds on GPU-based providers finish in under 5 seconds on the WSE-3. Paid tier limits are not publicly documented.
SambaNova
Hardware: Custom RDU (Reconfigurable Dataflow Unit). Best time-to-first-token (TTFT): approximately 0.2 seconds.
| Tier | RPM | Notes |
|---|---|---|
| Free | 10–30 (varies by model) | Hosts up to Llama 4 Maverick for free |
SambaNova offers Llama 4 Maverick (1M context) on the free tier — most other free-tier providers cap out at smaller models. TPM and TPD limits are not publicly documented.
Audio Models — ASH & ASD
Groq and Fireworks both publish audio-specific limits.
| Provider / Model | RPM | RPD | ASH | ASD | Audio min/min |
|---|---|---|---|---|---|
| Groq (Free) — Whisper Large v3 | 20 | 2,000 | 7,200 | 28,800 | — |
| Groq (Free) — Whisper Large v3 Turbo | 20 | 2,000 | 7,200 | 28,800 | — |
| Fireworks AI — Whisper v3-large | — | — | — | — | 200 |
| Fireworks AI — Whisper v3-turbo | — | — | — | — | 400 |
Groq: 7,200 ASH = 2 hours of audio per hour of wall time. 28,800 ASD = 8 hours per day. Adequate for a podcast transcription pipeline on the free tier.
Fireworks: 200 min/min for Whisper v3-large, 400 min/min for v3-turbo. Concurrent streaming capped at 10 connections.
OpenAI’s GPT Transcribe and GPT Live Transcribe are available via API. Independent benchmarks show both lag ElevenLabs, Google, and Mistral on error rates. OpenAI does not yet publish audio-specific ASH/ASD figures for these endpoints in its public rate limit documentation.
Cloud Aggregators — Azure AI & AWS Bedrock
Both use configurable per-deployment limits, not fixed tiers. The shared ratio is 6 RPM per 1,000 TPM.
Azure AI (Microsoft)
| Deployment Type | How Limits Work |
|---|---|
| Pay-as-you-go (Standard) | TPM quota per model per region, auto-scales with usage |
| Provisioned (PTU) | Reserved throughput units, no per-request limits |
| Global/Data Zone | Higher default quotas, multi-region routing |
GPT-5.6 Sol remains the default in Microsoft 365 Copilot. Azure AI hosts GPT-6 Astra, GPT-6.1 Sol, GPT-5.6 Sol, GPT-5.5, GPT-5.4, Claude Fable 5.1/Opus 5.5/Sonnet 5.5, and Gemini 3.1 Pro across regions. Gemini 4 Argon, 3.8 Flash, 3.7 Flash, and 3.6 Flash availability on Azure is not yet confirmed — verify model and region availability before planning deployments. PTU deployments remove per-request limits. Azure AI Quotas & Limits
AWS Bedrock
Hosts Claude (Fable 5.1, Opus 5.5, Sonnet 5.5, Haiku 5.5, and legacy models), Llama 4, Mistral, and Nova models with per-model, per-region quotas. Amazon is scaling back most Nova AI models (Premier, Omni, Reel, Canvas into “keep the lights on” mode) while a new Frontier research team works on a foundation model expected at re:Invent. Default quotas vary by model and can be increased via AWS Service Quotas console. Provisioned Throughput deployments remove rate limits entirely. AWS Bedrock Quotas
More Providers — Perplexity, Alibaba (Qwen), Moonshot (Kimi)
| Provider / Model | RPM (T0) | RPM (T1) | RPM (T3+) | TPM | TPD (T0) |
|---|---|---|---|---|---|
| Perplexity — Sonar Pro | 50 | 150 | 1,000 | — | — |
| Perplexity — Sonar | 50 | 150 | 1,000 | — | — |
| Perplexity — Deep Research | 5 | 10 | 40 | — | — |
| Alibaba (Qwen) — Qwen3.8-Max* | — | 600* | 600* | 1M* | — |
| Alibaba (Qwen) — Qwen3.7-Max† | — | 600† | 600† | 1M† | — |
| Alibaba (Qwen) — Qwen3.5 Plus | — | 15,000 | 30,000 | 5M | — |
| Alibaba (Qwen) — Qwen3.5 Flash | — | 15,000 | 30,000 | 10M | — |
| Moonshot — Kimi K3‡ | 3‡ | 200‡ | 5,000‡ | — | 1.5M‡ |
| Moonshot — Kimi K2.5 | 3 | 200 | 5,000 | — | 1.5M |
*Qwen3.8-Max rate limit figures not yet published. Qwen3.7-Max figures carried forward as a provisional floor. Qwen3.8-Max per-token pricing is not published — do not use any per-token figure for it. †Qwen3.7-Max rate limit figures not yet formally published. Qwen3 Max limits carried forward as a provisional floor. ‡Kimi K3 rate limit figures not yet confirmed in Moonshot’s public documentation. K2.5 figures carried forward as a provisional floor.
Perplexity tiers: T0 = new account, T1 = $50+, T3 = $500+. Alibaba limits shown for international (Singapore) deployment; Beijing deployment is more generous (Qwen3.5 Flash at 30K RPM — Qwen3.8-Max and 3.7-Max Beijing limits not yet confirmed). Moonshot T1 = $10+ cumulative recharge; T0 has 1.5M TPD cap, T1+ is unlimited.
Standout: Alibaba’s Qwen3.5 Flash at 30K RPM / 10M TPM (Beijing) remains the highest confirmed throughput of any provider in this post. Moonshot’s T5 tier (10K RPM, 1,000 concurrent connections, $3,000+ recharge) is competitive at the high end. Kimi K3’s 2.8T parameters and Opus-class performance at Sonnet-level pricing make it a meaningful option for high-volume workloads — confirmed rate limit figures are still absent from Moonshot’s public documentation.
Provider Notes
- OpenAI: GPT-6 Astra (September 2026;
gpt-6-astra; 1,050,000 context, 128K max output; $10/$50 per M — prompts over 272K input tokens bill at 2x input/cache and 1.5x output for the whole request; Fast mode up to 2x speed at 2x price) is the current flagship. GPT-6.1 Sol (September 29;gpt-6.1-sol; $2/$10 per M, cached $0.10; nearly matches Astra on agentic coding, computer use, and professional work at approximately one-fifth of Astra’s token prices). GPT-6 Luna (September 22;gpt-6-luna; $0.10/$0.50 per M, cached $0.01; OpenAI’s cheapest current model and the stated replacement for the GPT-4.1 family and GPT-5.4 Nano). Rate limit tier tables for GPT-6 Astra, GPT-6.1 Sol, and GPT-6 Luna not yet published — GPT-5.5 figures carried forward as provisional floors. GPT-5.6 Sol repriced to $4/$20 per M (was $5/$30) after the GPT-6 launch; GPT-5.6 Terra $2/$12, GPT-5.6 Luna $0.20/$1.20. GPT-5.5 remains available ($5/$30 per M; variants: GPT-5.5 Thinking, GPT-5.5 Pro). GPT-5.5 Instant remains available. GPT-5.4 remains available ($2.50/$15 per M; variants: GPT-5.4 Thinking, GPT-5.4 Pro, GPT-5.4-Cyber for vetted security teams). GPT-5.4 Mini ($0.75/$4.50 per M) is still available. GPT-5.3 Codex consolidated into ChatGPT as a unified superapp interface; still active. GPT-5.2 being phased out. DEPRECATED: GPT-4.1 and GPT-4.1 Mini deprecated October 14, 2026, shut down January 14, 2027. GPT-4.1 Nano deprecated April 22, 2026, shuts down October 23, 2026 — days away. GPT-5.4 Nano deprecated October 1, 2026, shuts down April 1, 2027. GPT-6 Luna is the replacement for all of these. GPT-4 and GPT-4o retired February 2026. GPT Transcribe and GPT Live Transcribe available via API; audio-specific rate limit figures not yet published. Assistants API deprecated August 2026 — migrate to the Responses API. Batch API offers 50% discount with higher queue limits. Docs - Anthropic: Current lineup as of October 2026 is Claude Fable 5.1 / Opus 5.5 / Sonnet 5.5 / Haiku 5.5 — all 1M context, 128K max output, June 2026 reliable knowledge cutoff. Anthropic’s guidance is to start with Opus 5.5 for most workloads and reach for Fable 5.1 for demanding reasoning and long-horizon agentic work. Claude Fable 5.1 (September 1;
claude-fable-5-1; $10/$50 per M, cache reads $0.25): top-capability tier; can identify software vulnerabilities in source code; biology safeguards intervene on benign requests 85% less often than Fable 5. Claude Opus 5.5 (September 22;claude-opus-5-5; $4/$20 per M, cache reads $0.20): recommended starting point; always-on adaptive thinking, default effort medium. Claude Sonnet 5.5 (September 28;claude-sonnet-5-5; $2/$10 per M, cache reads $0.10): current balanced tier; adaptive thinking, default effort high. Claude Haiku 5.5 (October 7;claude-haiku-5-5): current fast/budget tier; tiered pricing: $0.10/$0.50 per M for prompts up to 100,000 tokens, $0.50/$2.50 above. Rate limit figures for all four not yet published — predecessor figures carried forward as provisional floors. Claude Mythos 5.1 (same underlying model as Fable 5.1; invitation-only via Project Glasswing — not generally available). Claude Fable 5 (June 9, 2026) and legacy models (Opus 5, Sonnet 5, Haiku 4.5, Opus 4.8) remain available. Sonnet 5 and Sonnet 5.5 pricing is $2/$10 per M permanently — the September 1 increase to $3/$15 was cancelled; $2/$10 is now the standard rate. Sonnet 4.6 is still $3/$15. Extended thinking (thinking.type "enabled"plusbudget_tokens) is deprecated on Opus 4.6 and Sonnet 4.6 and is NOT accepted on later models — use theeffortparameter for adaptive thinking on current models. The ~30%-more-tokens tokenizer arrived with the Claude 4.7 generation (not Sonnet 5): Claude 4.7 and later use it; Sonnet 4.6 and earlier use the previous one. Cached input tokens don’t count toward ITPM on current 4.x and 5.x models. No free tier; lowest entry is $5. Claude 3 Haiku retired April 2026. Docs - Google: Static rate limit tables removed from public docs in Q1 2026. Actual limits only visible in AI Studio dashboard (
aistudio.google.com/rate-limit). Gemini 4 Argon (October 13, 2026;gemini-4-argon) is the current flagship — 1M unified context, up to 1M output tokens with long decode (65,536 default), ranked #1 on the Vals Index; introductory $2/$10 per M through November 30, 2026, doubling January 1, 2027; larger-prompt tier at $4/$20 but exact threshold unconfirmed. Paid tier only. Rate limit figures not yet published. Gemini 3.8 Flash (September 2, 2026;gemini-3.8-flash): most intelligent Flash model, GA; 1M context, 64K max output, tunable thinking levels; introductory $0.75/$3.75 per M through December 31, 2026, rising to $1.50/$7.50 January 1, 2027; rate limit figures not yet published. Gemini 3.8 Flash Cyber (cybersecurity domain) launched alongside it. Gemini 3.7 Flash (August 2026; previous flagship, still GA, still the default for Antigravity; same introductory pricing; rate limit figures not yet published). Gemini 3.6 Flash (July 2026; still GA; same introductory pricing; rate limit figures not yet published). Gemini 3.5 Flash (GA since May 19; ~4x faster output; rate limit figures not yet published). Gemini 3.5 Flash-Lite and Gemini 3.5 Flash Cyber launched July 2026; rate limit figures not yet published. Gemini 3.5 Pro remains in limited preview (2M context, Deep Think). Gemini 3.1 Pro remains available as the previous-generation Pro model ($2/$12 per M ≤200K, $4/$18 >200K). Gemini 2.0 Flash/Flash-Lite shut down June 1, 2026. Docs - Groq: Free tier publishes exact numbers (30 RPM, 6K–30K TPM depending on model). Daily token budgets (TPD) alongside TPM. Cached tokens don’t count. Hosts Llama 4 Scout and other open models on free tier. Runs on custom LPU hardware. Docs
- xAI (SpaceXAI): Current flagship is Grok 4.7 (September 21, 2026;
grok-4.7; 500K context; $2/$6 per M below 200K prompt tokens, $4/$12 above; knowledge cutoff May 2026). Grok 4.6 is the previous flagship, still available ($2/$6 per M, same context and pricing structure; knowledge cutoff February 1, 2026). Grok 4.5 (~July 9, 2026) older still.grok-build-0.1is the agentic coding model ($1/$2 per M). 5-tier structure ($0/$50/$250/$1K/$5K thresholds) but numerical RPM/TPM only visible in xAI Console. Docs - Mistral: European data residency (GDPR). 5-tier structure (Free through Tier 4 at $500+). Limits enforced per RPS (not RPM), TPM, and tokens/month. Actual numbers require login at
admin.mistral.ai. Mistral Large 4 (October 6, 2026;mistral-large-4; 1.05T total / 52B active plus 1.6B vision encoder; 1M context; public preview, not GA; per-token pricing not confirmed — do not quote one; licence not confirmed). Mistral Medium 3.5 ($1.50/$7.50 per M) is the proprietary frontier tier. Mistral Large 3 (675B MoE, Apache 2.0, $0.50/$1.50 per M) and Mistral Small 4 (Apache 2.0, $0.15/$0.60 per M) are available. Devstral 2 (72.2% SWE-bench, $0.40/$2.00 per M) available for coding workloads. Docs - DeepSeek: Fully dynamic concurrency limits based on server load. No published RPM/TPM. DeepSeek-V4.1-Flash (
deepseek-flash) is the current flagship — 552B MoE with 8B active for input and 16B for output; native multimodal visual understanding; 1M context, 384K max output; pricing varies peak vs. off-peak: $0.30/$1.20 per M at peak, $0.15/$0.60 at off-peak (peak hours: 01:00–04:00 and 06:00–10:00 UTC, Monday–Friday, excluding Chinese public holidays) — always specify which rate you are quoting. The olderdeepseek-v4-flashanddeepseek-v4-flash-vision-expnames are retired and route to V4.1-Flash. DeepSeek V4-Pro pricing is uncertain — a September 2026 announcement says V4-Pro traffic reroutes to V4.1-Flash, but the live pricing page still lists V4-Pro separately; check both sources. V3.2 aliases deprecated July 24, 2026 — those endpoints are gone. Docs - Cerebras: Fastest inference (approximately 2,100 tok/s on Llama 4 Scout). Free tier: 30 RPM, 60K TPM, 1M TPD. Paid tier limits not publicly documented. Custom WSE-3 silicon. Docs
- SambaNova: Best TTFT (approximately 0.2s). Free tier: 10–30 RPM depending on model size. Hosts up to Llama 4 Maverick for free. TPM and TPD limits not publicly documented. Custom RDU hardware. Docs
- Together AI: Dynamic rate limits since January 2026. No fixed tiers or published numbers. Limits grow with sustained usage and are returned in API response headers. Docs
- Fireworks AI: Dynamic ceiling up to 6,000 RPM (soft limit starts approximately 1 RPS, doubles hourly). On-demand GPU deployments remove limits. Spending tier caps: $50–$50K/month by tier. Docs
- Cohere: Command A+ (May 20, 2026; Cohere’s first MoE, 218B total / 25B active; 128K input context and 64K max output; vision input, reasoning, translation, and agentic tasks; native citations; runs on 1x B200 or 2x H100; 48 languages including all official EU languages) is the flagship. Licence is unconfirmed — do not state Apache 2.0 or any other licence without checking Cohere’s model page. Self-hosting removes API rate limits entirely. API rate limit figures for Command A+ not yet published — 500 RPM (production chat) shown in tables is the Command R+ figure. Trial keys: 1,000 calls/month. No per-token API price published for Command A+. Docs
- Perplexity: 6 tiers (T0–T5) based on cumulative spend. Leaky bucket algorithm. Deep Research model has very low limits (5–100 RPM). Agent API: 50–2,000 RPM by tier. Docs
- Alibaba (Qwen): Qwen3.8-Max (August 3, 2026; 2.4T total / 95B active sparse MoE with hybrid attention; 1M context; 5th on Text Arena, 2nd on Vision Arena; weights now released as Qwen3.8-2.4T-A95B on Hugging Face and ModelScope — the first Qwen-Max-class model to ship openly) is the current flagship. Published per-token pricing not yet confirmed — write “not published” rather than guessing. Qwen3.8-27B (open-weight native multimodal dense, runs on consumer hardware) and Qwen3.8-Flash also released. Qwen3.7-Max (May 20; $2.50/$7.50 per M) is the previous flagship, still available. Open-weight family includes Qwen3.6 (Apache 2.0 — e.g., Qwen3.6-35B-A3B, Qwen3.6-27B). Region-specific limits apply — Beijing deployment is more generous than Singapore/Global. Qwen3.5 Flash at 30K RPM / 10M TPM (Beijing) is the highest confirmed throughput of any provider listed. Docs
- Moonshot (Kimi): 6 tiers (T0–T5). T0 ($1 recharge): 3 RPM, 1 concurrent, 1.5M TPD. T5 ($3,000): 10K RPM, 1,000 concurrent, 5M TPM, unlimited TPD. Kimi K3 (2.8T parameters, 1M context, Opus-class performance at Sonnet-level pricing) is the current flagship — rate limit figures not yet confirmed in public documentation; K2.5 figures carried forward as a provisional floor. Automatic 75% caching discount applied without opt-in. Docs
- AWS Bedrock: Hosts Claude (Fable 5.1, Opus 5.5, Sonnet 5.5, Haiku 5.5, and legacy Claude models), Llama 4, Mistral with per-model per-region quotas. Amazon scaling back most Nova AI models (Premier, Omni, Reel, Canvas) while a new Frontier team works on a foundation model expected at re:Invent. 6 RPM per 1K TPM ratio. Provisioned Throughput removes limits. Docs
- Azure AI: Hosts GPT-6 Astra, GPT-6.1 Sol, GPT-5.6 Sol (default in Microsoft 365 Copilot), GPT-5.5, GPT-5.4, Claude Fable 5.1/Opus 5.5/Sonnet 5.5, and Gemini 3.1 Pro across regions. Gemini 4 Argon, 3.8 Flash, 3.7 Flash, and 3.6 Flash availability on Azure not yet confirmed — verify model and region availability before planning deployments. PTU deployments remove per-request limits. Docs
- NVIDIA NIM: Hosted API is for prototyping (40 RPM). Self-hosted NIM containers have no limits. Docs
Tips for managing rate limits
For a deep dive on token pricing, cost optimization strategies, caching mechanics, model routing architectures, and production cost modeling, see LLM Token Costs and Efficiency. This section covers rate limit management specifically.
Handling 429s
Use exponential backoff when you hit 429s. The Anthropic SDK, OpenAI SDK, and most third-party clients handle this automatically. Don’t build your own retry loop unless you need custom jitter or circuit-breaking behavior.
Check response headers before you hit the wall. Together AI and Fireworks return your current rate limit state in every API response. DeepSeek adjusts concurrency limits dynamically based on server load. Reading these headers in production enables proactive throttling rather than reactive retries.
GPT-4.1 Nano shutdown is imminent
GPT-4.1 Nano shuts down October 23, 2026 — days from this edition’s publish date. Any pipeline referencing gpt-4.1-nano will stop functioning at that cutoff. GPT-6 Luna ($0.10/$0.50 per M) is OpenAI’s named replacement. GPT-5.4 Mini ($0.75/$4.50 per M) is an intermediate option if Luna is too stripped-down for a given workload. Migrate now.
GPT-6 Astra, GPT-6.1 Sol, and GPT-6 Luna: treat published limits as provisional
All three models are broadly available but their rate limit tier tables are not yet published. Figures in this post are carried forward from GPT-5.5. GPT-6 Astra runs $10/$50 per M; GPT-6.1 Sol runs $2/$10 per M; GPT-6 Luna runs $0.10/$0.50 per M. Check platform.openai.com/account/limits for your account’s actual current limits.
GPT-5.6 Sol repricing: update cost models
GPT-5.6 Sol dropped from $5/$30 to $4/$20 per M after the GPT-6 launch. Any cost model carrying the old $5/$30 figure is now wrong. GPT-5.5 is unchanged at $5/$30. The “up to 80% cut” OpenAI marketing language describes movement across the Sol/Terra/Luna spectrum — do not derive per-token costs from that claim; use the verified figures.
Current Anthropic models: revise tokenizer math
The ~30%-more-tokens tokenizer arrived with the Claude 4.7 generation, not with Sonnet 5. Claude 4.7 and later all use it — that includes the entire current lineup (Fable 5.1, Opus 5.5, Sonnet 5.5, Haiku 5.5). A workload that consumed 20K ITPM on Sonnet 4.6 will consume approximately 26K ITPM on any current Claude model with the same inputs. Recheck limits against actual token counts, particularly at Tier 1 where ITPM is constrained at 30K (provisionally carried figure).
Anthropic adaptive thinking: no budget_tokens on current models
Extended thinking (thinking.type "enabled" plus budget_tokens) is deprecated on Opus 4.6 and Sonnet 4.6 and is NOT accepted on current models. Current models use adaptive thinking steered by the effort parameter. Code samples targeting Fable 5.1, Opus 5.5, Sonnet 5.5, or Haiku 5.5 must not include budget_tokens.
Gemini introductory pricing: two separate expiry dates
Gemini 4 Argon’s introductory pricing ($2/$10 per M) expires November 30, 2026 — rates double January 1, 2027. Gemini 3.8 Flash, 3.7 Flash, and 3.6 Flash all share a separate introductory window ($0.75/$3.75) expiring December 31, 2026 — standard $1.50/$7.50 from January 1, 2027. Cost models projecting past either date should use the standard rates.
Caching for throughput (not just cost)
Prompt caching has a rate limit benefit separate from its cost benefit. The cost savings are covered in the token costs post. The rate limit implications:
- Anthropic: Cached input tokens don’t count toward ITPM limits at all on Claude 4.x and 5.x models. With an 80% cache hit rate, a 2M ITPM limit effectively handles approximately 10M total input tokens per minute. Cache read multipliers differ by model: 0.025x input on Fable 5.1 ($0.25/M), 0.05x on Opus 5.5 ($0.20/M) and Sonnet 5.5 ($0.10/M), and 0.10x on everything else. Compute cache read costs from the verified input price rather than assuming a flat 10%.
- Groq: Cached tokens don’t count toward TPM/TPD limits on the free tier, effectively expanding your daily token budget.
- Moonshot (Kimi): Automatic 75% caching discount applied without opt-in. No cache management required.
DeepSeek V3.2 is gone; V4-Pro routing is uncertain
The July 24 deprecation deadline for V3.2 aliases has passed. If anything in your pipeline references a V3.2 alias, it is receiving errors. deepseek-flash is the current endpoint for V4.1-Flash. V4-Pro routing is uncertain — a September 2026 announcement says V4-Pro requests now route to V4.1-Flash, but the pricing page still lists V4-Pro separately. Check both sources before assuming V4-Pro pricing.
DeepSeek peak vs. off-peak pricing
DeepSeek-V4.1-Flash pricing varies by time of day. Peak ($0.30/$1.20 per M) applies during 01:00–04:00 and 06:00–10:00 UTC, Monday–Friday, excluding Chinese public holidays. Off-peak ($0.15/$0.60) applies at all other times. Always specify which rate you are quoting in cost models.
OpenAI Assistants API is deprecated
Deprecated August 2026. The Responses API is the replacement. Migration guidance is in OpenAI’s migration docs.
Batch APIs: higher queue limits
OpenAI, Anthropic, and Google all offer batch endpoints with queue limits that far exceed real-time TPM. For workloads that don’t need sub-second responses, batch endpoints let you move more tokens through tighter rate limits. See batch API cost savings for pricing details.
Model routing for rate limit distribution
Routing requests to different models by complexity distributes rate limit pressure across multiple buckets instead of concentrating it on your most constrained tier. A three-tier architecture (budget for triage, mid-tier for generation, flagship for reasoning) also cuts costs substantially. Full routing architecture with cost modeling is in the token costs post.
GPT-6 Luna ($0.10/$0.50 per M) and DeepSeek V4.1-Flash (from $0.15/$0.60 per M off-peak) are strong candidates for the triage and generation layers of a routing stack. Kimi K3 (Opus-class at Sonnet-level pricing) is worth evaluating as a reasoning layer once rate limit figures are confirmed. Claude Haiku 5.5 ($0.10/$0.50 per M for prompts under 100K tokens) is roughly 75% cheaper to run than Haiku 4.5 and is designed specifically for high-volume, latency-sensitive work like classification, extraction, and routing.
Custom silicon for fewer concurrent connections
Cerebras (approximately 2,100 tok/s) and SambaNova (approximately 0.2s TTFT) are faster than GPU-based providers by a large margin. A task that requires 10 parallel GPU-based API calls to meet a latency SLA might need 2–3 calls on Cerebras. Fewer concurrent calls means less rate limit pressure per unit of output.
Self-hosting as an escape hatch
Cohere’s Command A+ (open-weight, 218B MoE / 25B active) runs on 2x H100 or a single B200. Qwen3.8-Max (2.4T parameters, weights now publicly released as Qwen3.8-2.4T-A95B) and Kimi K3 (2.8T parameters) are also available for self-hosting at substantially higher hardware requirements. Mistral Large 4 (1.05T total / 52B active, open-weight, public preview) and Thinky Inkling (975B, Apache 2.0) round out the open-weight options. For teams already operating that hardware, self-hosting removes API rate limits entirely. The economics only work at scale — hardware cost needs to be weighed against API spend.
Watch for shared pools and hidden constraints
- Anthropic model pools: Fable 5.1 pool assignment is not yet formally documented. Opus 5.5 pool assignment not yet