AI pricing & rate-limit changelog
Every recorded change, newest first: 70 dated entries from official sources across 16 companies. Each row shows the old and new values as stated by the source, a link to the official page, and the company page. New entries are added to the top automatically at each build. Follow by RSS or JSON Feed.
Backfilled = added after the fact from archived copies of the official page, not caught live; its date is the narrowest window the archives support ('between X and Y'), or the month when the exact day isn't stated.
| Date | Change | |||||
|---|---|---|---|---|---|---|
| 2026-10-22 | Anthropic (Claude) | quota | Investigating User reports that, since Claude Opus 5.5, weekly allowances on Claude subscriptions run out faster than a few weeks earlier with the same workflows, and asks Anthropic to explain any change. | user report | not confirmed | public post |
| 2026-10-20 | OpenAI (ChatGPT) pricing 12-hour prompt caching option added; 12-hour cache writes cost 2.5x the input token price.Old No 12-hour caching option (not stated)New 12-hour caching available; writes at 2.5x input price | |||||
| 2026-10-18 | Fireworks AI rate-limit Small rate-limit tier threshold raised from <600B to <800B total parameters; ceilings unchanged.Old Small tier: fewer than 600B parametersNew Small tier: fewer than 800B; Medium: 800B to <1.6T | |||||
| 2026-10-15 | Anthropic (Claude) rate-limit Fast mode for Claude Opus 5.5 left research preview and received a default rate limit in Start, Build and Scale tiers.Old Research preview, access by requestNew Default fast mode rate limit in Start, Build, and Scale tiers; no beta header needed | |||||
| between 2026-09-28 and 2026-10-13 Backfilled | xAI (Grok) rate-limit Per-second request limit formula loosened from RPM / 60 to RPM / 48 with a minimum of 2.Old RPS = RPM / 60New RPS = RPM / 48, minimum 2 | |||||
| between 2026-07-08 and 2026-10-12 Backfilled | Amazon Bedrock (AWS) quota Quotas page now states applied quotas may be lower than listed defaults based on account age, usage and payment history, and increases aren't automatic.Old Adjustment factors listed without account age or usage historyNew Applied quotas may be lower than defaults (account age, usage history, payment history, fraud signals) | |||||
| 2026-10-08 | GitHub Copilot quota Org owners can bill Copilot code reviews to the organization instead of consuming members' Copilot quotas.Old Billed to member's own Copilot entitlement only; review fails if quota exhaustedNew Option to bill the repository-owning organization (requires AI Credits paid usage) | |||||
| 2026-10-07 Backfilled | Hugging Face quota Free users no longer receive included monthly Inference Providers credits.Old Free users: $0.10 monthly credits, subject to changeNew Free users: none (purchase credits or upgrade to PRO) | |||||
| 2026-10-07 | Fireworks AI rate-limit Adaptive rate-limit ceilings for Small serverless models doubled.Old Total Prompt TPM 108M; Uncached 27M; Generated 1.08M (per 2026-10-02 entry)New Total Prompt TPM 216M; Uncached 54M; Generated 2.16M | |||||
| 2026-10-07 | Anthropic (Claude) plan Max and Team plans now include monthly API credits for the Claude Platform.Old No included API credits (not stated)New Monthly API credits included; claimed by linking a Claude Console organization | |||||
| 2026-10-07 | Anthropic (Claude) pricing Prompt cache read price on Claude Sonnet 5.5 lowered.Old $0.20 per MTok (0.1x base input)New $0.10 per MTok (0.05x base input)$0.20 → $0.10 per MTok | |||||
| 2026-10-06 | OpenAI (ChatGPT) rate-limit API usage tiers simplified from five to three (Build, Launch, Grow).Old Five usage tiersNew Three usage tiers: Build, Launch, Grow | |||||
| 2026-10-02 | Together AI rate-limit Dynamic serverless rate limits that scaled with recent traffic were retired.Old Dynamic rate limits scaling with recent trafficNew No dynamic rate limits; requests may still be limited during high demand | |||||
| 2026-10-02 | Fireworks AI rate-limit Adaptive rate-limit ceilings for Small serverless models increased.Old Previous value not recordedNew Total Prompt TPM 108M; Uncached Prompt TPM 27M; Generated TPM 1.08M | |||||
| Oct 2026 (exact day not stated) Backfilled | Perplexity pricing Decisions API model replaced by pplx-decider-v1.1-27b with lower input price (changelog dated October 2026, day not stated).Old $0.04 per 1M input tokensNew $0.02 per 1M input tokens$0.04 → $0.02 per 1M input tokens | |||||
| Oct 2026 (exact day not stated) Backfilled | Perplexity pricing Agent API low, medium, high and xhigh presets switched to Fast Search; web_search price lowered (changelog dated October 2026, day not stated).Old $2.50 per 1,000 web_search invocationsNew $1.00 per 1,000 invocations$2.50 → $1.00 per 1,000 invocations | |||||
| 2026-10-01 | Fireworks AI pricing Serverless price for DeepSeek V4.1 Flash increased (uncached input / cached input / output per 1M).Old Standard $0.22/$0.007/$0.66; Priority $0.275/$0.00875/$0.825New Standard $0.30/$0.006/$1.20; Priority $0.375/$0.0075/$1.50 | |||||
| 2026-09-29 | Windsurf plan Users can sign in with ChatGPT and bill GPT model usage to an eligible ChatGPT Plus or Pro plan.Old Previous value not recordedNew GPT model requests billable to eligible ChatGPT Plus/Pro plan | |||||
| 2026-09-29 | Amazon Bedrock (AWS) quota Per-model tokens-per-day quota replaced by a cross-model, per-account, per-Region daily token quota.Old Per-model 'Model invocation max tokens per day' quotaNew 'Cross-Model Max Tokens Per Day' quota per account per Region | |||||
| between 2026-09-10 and 2026-09-28 Backfilled | xAI (Grok) rate-limit Voice models brought under the spend-based rate limit tiers.Old Tiers apply to text and embedding models; Voice increases via contactNew Tiers apply to text, embedding, and voice models | |||||
| 2026-09-24 | Anthropic (Claude) pricing Billing resumed for refusals before any output in the bio, frontier_llm and reasoning_extraction categories.Old Pre-output refusals in these categories not billedNew Billed at the rates of the model that ran it; other categories still not billed | |||||
| between 2026-09-22 and 2026-09-23 Backfilled | Ollama pricing Off-peak per-token pricing introduced for cloud models.Old Single per-token rateNew Off-peak rates outside 12:00-18:00 UTC weekdays and all weekend; off-peak amounts per model not reproduced here (see capture) | |||||
| 2026-09-23 | Cursor (Anysphere) plan Time-limited (10-day) included usage credits for the new Rollouts bot on Teams and Enterprise.Old No Rollouts (new product)New Credits for roughly 50 (Teams) / 500 (Enterprise) changes for 10 days | |||||
| 2026-09-22 | Together AI pricing Serverless prices lowered for Qwen3.7-Max and Qwen3.8-Flash.Old Qwen3.7-Max $2.50/$7.50; Qwen3.8-Flash $0.15/$0.47 per 1M (in/out)New Qwen3.7-Max $1.50/$4.50; Qwen3.8-Flash $0.09/$0.282 per 1M (in/out)$2.50 → $1.50 · $7.50 → $4.50 · $0.15 → $0.09 · $0.47 → $0.28 | |||||
| 2026-09-22 | Anthropic (Claude) quota Five-hour usage limits increased on Pro, Max, Team and seat-based Enterprise plans at the Claude Opus 5.5 launch; subscription users also got a saveable rate limit reset. Weekly limits are not mentioned in the announcement.Old Previous value not recordedNew Five-hour usage limits increased (size of increase not stated); one saveable rate limit reset for subscription users | |||||
| 2026-09-02 | xAI (Grok) pricing grok-imagine-image-quality retires 2026-11-02; requests will be served by grok-imagine-image-2.0 (quality=low) at a lower per-image price.Old grok-imagine-image-quality at its per-image priceNew Routed to grok-imagine-image-2.0 quality low, lower per-image price | |||||
| Sep 2026 (exact day not stated) Backfilled | Perplexity pricing Agent API fast preset switched to Fast Search; web_search price per 1,000 invocations lowered (changelog dated September 2026, day not stated).Old $2.50 per 1,000 web_search invocationsNew $1.00 per 1,000 invocations$2.50 → $1.00 per 1,000 invocations | |||||
| 2026-09-01 | Fireworks AI rate-limit Serverless adaptive rate-limit ceilings now vary by model size tier.Old Previous value not recordedNew Small models higher ceilings, medium intermediate, large lower | |||||
| 2026-08-31 Backfilled | Ollama plan Ollama Cloud plans moved to per-token pricing with monthly included usage credits, replacing 5-hour and weekly limits; Team plan launched.Old Session limits resetting every 5 hours and weekly limits (per Wayback capture of ollama.com/pricing)New Pro $20/mo incl. $60 usage; Max $100/mo incl. $300; Team $500/mo incl. $1,000 shared; no 5-hour or weekly limits | |||||
| 2026-08-31 | GitHub Copilot plan Model access for users with seats in several organizations is now determined only by the billing organization.Old Model usable if any of the user's organizations enabled itNew Billing organization decides model access | |||||
| 2026-08-28 | GitHub Copilot plan Credit-card/PayPal Copilot Business and Enterprise seats move to upfront per-seat charges; prices unchanged.Old Seats not charged upfront (not stated)New Upfront charge per assigned seat at start of billing cycle; existing customers from 2026-10-01 | |||||
| 2026-08-27 Backfilled | Perplexity pricing Agent/Orchestrator API price for google/gemini-3.7-flash increased.Old Previous value not recordedNew $0.75 / 1M input, $0.075 / 1M cached input, $3.75 / 1M output and reasoning | |||||
| 2026-08-21 | OpenAI (ChatGPT) pricing GPT-5.6 Sol price reduced to $4 input / $20 output per 1M tokens (promotional, at least through 2026-11-21).Old Previous value not recordedNew $4 / 1M input, $20 / 1M output | |||||
| 2026-08-21 Backfilled | Hugging Face plan Monthly credits included with PRO, Team and Enterprise became general-purpose compute credits usable beyond Inference Providers.Old $2/month (PRO) and $2/seat/month credits for Inference ProvidersNew Same amounts, spendable on all Hugging Face compute services | |||||
| 2026-08-10 | Anthropic (Claude) pricing Claude Sonnet 5 introductory price made standard; scheduled increase to $3/$15 on 2026-09-01 cancelled.Old $2 / $10 per MTok introductory, scheduled to rise to $3 / $15 on 2026-09-01New $2 / $10 per MTok standard | |||||
| 2026-07-30 | OpenAI (ChatGPT) plan Fast mode replaced the Priority Processing offering; for GPT-5.6 Sol it is priced at twice standard.Old Priority Processing tierNew Fast mode, up to 2.5x faster at twice the price (GPT-5.6 Sol); priority-tagged requests auto-use Fast mode | |||||
| 2026-07-30 | OpenAI (ChatGPT) pricing GPT-5.6 Luna price cut 80% and GPT-5.6 Terra price cut 20%.Old Previous value not recordedNew Luna: 80% lower; Terra: 20% lower | |||||
| between 2026-07-28 and 2026-07-29 Backfilled | Ollama plan Team plan listed: per-seat pricing with five-seat minimum.Old No Team plan listedNew $25/seat/month, minimum 5 seats ($125/month) | |||||
| 2026-07-28 Backfilled | Cursor (Anysphere) plan New Cursor Start plan for developers in India at INR 649 per month, billed in INR via UPI or card.Old No India-specific plan (not stated)New Cursor Start: ₹649/month tax inclusive | |||||
| 2026-07-21 Backfilled | Vercel AI Gateway pricing AI Gateway added service tiers (default, priority, flex) for OpenAI and Gemini models, billed per tier actually used.Old Single default tier (not stated)New priority ~1.8-2x default pricing; flex ~0.5x default pricing | |||||
| between 2026-06-15 and 2026-07-06 Backfilled | xAI (Grok) rate-limit Request limits switched from requests per minute to requests per second, derived as RPM / 60.Old RPM and TPM caps per modelNew RPS (= RPM / 60) and TPM caps per model | |||||
| Jul 2026 (exact day not stated) Backfilled | Perplexity pricing Agent API prices for GPT-5.6 Luna and Terra lowered; GPT-5.6 Sol Fast mode added at 2x standard token pricing (changelog dated July 2026, day not stated).Old Previous value not recordedNew Luna $0.20/$1.20 per 1M in/out; Terra $2/$12; Sol Fast mode 2x | |||||
| 2026-06-30 Backfilled | Amazon Bedrock (AWS) rate-limit AWS updated its RPM quota guidance: one endpoint has no per-model RPM quotas; on bedrock-runtime RPM quotas are model-specific and some Claude Opus models have none.Old Previous value not recordedNew Some models, including Claude Opus 4.7 and 4.8, not subject to an RPM quota | |||||
| between 2026-06-06 and 2026-06-28 Backfilled | Vercel AI Gateway plan Free tier ($5/month credit) limited to a subset of models with lower per-model rate limits than paid tier.Old Free credits usable across all modelsNew Free tier: subset of Free Tier eligible models, lower per-model rate limits | |||||
| between 2026-06-23 and 2026-06-27 Backfilled | Google (Gemini) rate-limit Spend-based rate limits per rolling 10-minute window added alongside RPM/TPM.Old No spend-based rate limit listedNew Per 10 minutes: Free N/A; Tier 1 $10; Tier 2 $200; Tier 3 $200 | |||||
| 2026-06-15 Backfilled | xAI (Grok) plan A priority service tier became available on text inference endpoints; priority rates are billed only when priority is applied.Old No per-request priority tier (not stated)New service_tier "priority" on Chat Completions and Responses; priority rates billed only when used (rate not stated in note) | |||||
| 2026-06-09 Backfilled | Together AI pricing DeepSeek-V4-Pro serverless price lowered; cached-input pricing added for GLM-5.1 and Qwen3.5-397B-A17B.Old DeepSeek-V4-Pro $2.10 input / $4.40 output per 1MNew DeepSeek-V4-Pro $1.74 / $3.48; GLM-5.1 cached $0.26; Qwen3.5-397B cached $0.35 per 1M | |||||
| between 2026-05-21 and 2026-06-02 Backfilled | DeepSeek pricing deepseek-v4-pro 75%-off promotional prices became the listed standard prices.Old Cache hit $0.003625 / miss $0.435 / output $0.87 per 1M shown as 75% off list $0.0145 / $1.74 / $3.48New $0.003625 / $0.435 / $0.87 per 1M as standard price (no discount label) | |||||
| between 2026-04-28 and 2026-05-29 Backfilled | Vercel AI Gateway pricing Pricing page lists charges for opt-in capabilities: Custom Reporting, team-wide provider allowlist, team-wide ZDR.Old No add-on charges listedNew Custom Reporting $0.075/1,000 writes, $5/1,000 queries; team-wide provider allowlist $0.10/1,000 successful requests; team-wide ZDR $0.10/1,000 requests (Pro and Enterprise) | |||||
| 2026-05-29 Backfilled | Together AI pricing Serverless prices increased for Qwen3.5-9B, Meta-Llama-3-8B-Instruct-Lite and Llama-3.3-70B-Instruct-Turbo.Old Qwen3.5-9B $0.10/$0.15; Llama-3-8B-Lite $0.10/$0.10; Llama-3.3-70B-Turbo $0.88/$0.88 per 1M (in/out)New Qwen3.5-9B $0.17/$0.25; Llama-3-8B-Lite $0.14/$0.14; Llama-3.3-70B-Turbo $1.04/$1.04 per 1M (in/out) | |||||
| 2026-05-27 Backfilled | Amazon Bedrock (AWS) quota AWS documented per-model input and output token quotas for the bedrock-mantle endpoint and how to request increases.Old Previous value not recordedNew Per-model input/output token quotas on bedrock-mantle (values in Service Quotas, not stated here) | |||||
| between 2026-05-14 and 2026-05-21 Backfilled | DeepSeek rate-limit Pricing page added a per-model concurrency limit row.Old No concurrency limit listedNew Concurrency limit 2500 (deepseek-v4-flash) and 500 (deepseek-v4-pro), per column order | |||||
| between 2026-05-12 and 2026-05-18 Backfilled | Ollama plan Pro and Max users can buy extra usage balance consumed after included plan limits.Old No extra usage option (not stated)New Extra usage balance drawn after included limits | |||||
| 2026-05-14 Backfilled | Together AI pricing Announced price increase for google/gemma-4-31b-it effective 2026-05-21.Old $0.20 input / $0.50 output per 1MNew $0.39 input / $0.97 output per 1M$0.20 → $0.39 · $0.50 → $0.97 | |||||
| 2026-04-24 Backfilled | Together AI pricing Billable training tokens for no-packing fine-tuning jobs now calculated as len(dataset) * max_seq_length.Old Previous value not recordedNew Training tokens = len(dataset) * max_seq_length for no-packing jobs | |||||
| 2026-04-22 Backfilled | Together AI rate-limit Build Tiers 1-5, Scale and Enterprise tier labels retired, dynamic rate limits enabled for all users, billing moved to fully prepaid.Old Build Tiers 1-5 / Scale / Enterprise tiers; model-specific tier gatesNew Dynamic rate limits; fully prepaid billing; $5 credit purchase is the only gate | |||||
| 2026-04-15 Backfilled | Windsurf plan Devin cloud agent included with every self-serve Windsurf plan, billed against existing quota and extra usage.Old Devin Cloud not included in Windsurf plans (not stated)New Included with self-serve plans; billed from quota/extra usage; up to $50 extra usage added for first Devin Cloud session | |||||
| 2026-04-15 Backfilled | Together AI pricing Serverless price increased for google/gemma-3n-E4B-it.Old $0.02 input / $0.04 output per 1MNew $0.06 input / $0.12 output per 1M$0.02 → $0.06 · $0.04 → $0.12 | |||||
| between 2026-04-02 and 2026-04-13 Backfilled | xAI (Grok) rate-limit Rate limit tiers consolidated from six to five (Tier 0-4) with new spend thresholds.Old Tier 1 $0; Tier 2 $50; Tier 3 $200; Tier 4 $500; Tier 5 $1,000; Tier 6 $5,000New Tier 0 $0; Tier 1 $50; Tier 2 $250; Tier 3 $1,000; Tier 4 $5,000 | |||||
| 2026-04-06 Backfilled | Windsurf pricing Adaptive model option launched for Pro, Max and Teams with promotional extra-usage token rates for two weeks.Old Previous value not recordedNew Extra usage beyond quota: $0.50 / 1M input, $2.00 / 1M output, $0.10 / 1M cache read (for 2 weeks) | |||||
| between 2026-03-29 and 2026-04-03 Backfilled | Google (Gemini) rate-limit Priority inference gets its own rate limits, defaulting to 0.3x the standard limit per model and tier.Old Previous value not recordedNew 0.3x the standard rate limit for each model and tier | |||||
| between 2026-03-21 and 2026-03-29 Backfilled | Google (Gemini) rate-limit Usage tier qualification lowered for Tier 2 and re-based on first payment for Tier 3; billing tier caps listed.Old Tier 2: total spend > $250 and 30 days since successful payment; Tier 3: total spend > $1,000 and 30 daysNew Tier 1 cap $250; Tier 2: paid $100 + 3 days from first payment (cap $2,000); Tier 3: paid $1,000 + 30 days (cap $20,000 - $100,000+) | |||||
| between 2026-03-13 and 2026-03-23 Backfilled | xAI (Grok) rate-limit Spend-based rate limit tiers introduced for text models, unlocking automatically by cumulative spend since Jan 1 2026.Old Higher limits by emailing xAINew Tier 1 $0; Tier 2 $50; Tier 3 $200; Tier 4 $500; Tier 5 $1,000; Tier 6 $5,000; Enterprise on request | |||||
| 2026-03-23 Backfilled | Google (Gemini) plan Prepay and Postpay billing plans rolled out in AI Studio; existing accounts may be affected.Old Previous value not recordedNew Prepay and Postpay billing plans | |||||
| 2026-03-19 Backfilled | Windsurf quota Windsurf editor added support for a new quota billing system with daily and weekly quota usage shown in the IDE.Old Previous value not recordedNew Quota billing with daily and weekly quotas (amounts not stated in changelog) | |||||
| between 2026-03-12 and 2026-03-16 Backfilled | Ollama rate-limit Pricing page began stating explicit session (5-hour) and weekly usage limits for every plan.Old Previous value not recordedNew Session limits reset every 5 hours; weekly limits reset every 7 days | |||||
| 2026-03-16 Backfilled | Google (Gemini) rate-limit Gemini API usage tiers revamped and Billing Account spend caps introduced.Old Previous usage tiers (not stated)New Revamped usage tiers (thresholds not stated in changelog); Billing Account spend caps | |||||
| 2026-03-12 Backfilled | Google (Gemini) quota Project-level spend caps added to billing in AI Studio.Old Previous value not recordedNew Project-level spend caps available (amounts user-set) | |||||
| 2026-02-25 Backfilled | Hugging Face plan Free Hugging Face users can now continue past monthly credits by purchasing credits (pay-as-you-go previously PRO/Team/Enterprise only).Old Free users: $0.10 monthly credits, no pay-as-you-goNew Free users: $0.10 monthly credits plus pay-as-you-go with purchased credits | |||||
| between 2026-02-02 and 2026-02-03 Backfilled | Google (Gemini) pricing Grounding with Google Search billing beyond the free allowance moved from 'Coming soon' to active for models listed with 1,500 free requests per day (models not identified in capture text).Old 1,500 RPD free, then (Coming soon) $14 / 1,000 search queriesNew 1,500 RPD free, then $14 / 1,000 search queries | |||||
| between 2026-01-25 and 2026-01-30 Backfilled | v0 (Vercel) pricing v0 model token prices raised and cache write pricing listed.Old v0 Mini $0.50/1M input, $2/1M output; Pro $1.50/$7.50; Max $2/$10New v0 Mini $1/1M input, $5/1M output; Pro $3/$15; Max $5/$25; cache write $1.25/$3.75/$6.25 and cache read $0.10/$0.30/$0.50 per 1M added |
Understanding the terms: How AI billing works · How Claude's weekly usage limits work · Monthly digests