Skip to main content
RapidDev - Software Development Agency
AI API Limits & Performance Matrix12 min readVerified July 10, 2026

Claude 3.5 Sonnet API Rate Limits, Pricing & Performance (July 2026)

Claude 3.5 Sonnet was dropped from Anthropic's active pricing on February 19, 2026 and is not recommended for new projects. Its successor, Claude Sonnet 5 (claude-sonnet-5), offers 1M token context, introductory pricing at $2.00 input / $10.00 output per MTok (through August 31, 2026), and up to ~50 RPM on the Start tier. Cache reads are excluded from ITPM, effectively multiplying throughput headroom.

4.9Clutch rating
600+Happy partners
17+Countries served
190+Team members

Deprecated model

Sunsets Not announced as of July 10, 2026 — legacy string remains callable but is not on Anthropic's active pricing record.

Claude 3.5 Sonnet was dropped from Anthropic's daily pricing record on February 19, 2026. It remains callable as a pinned legacy API string but is not recommended for new projects. Anthropic's current recommended default for price/performance is Claude Sonnet 5 (claude-sonnet-5), released June 30, 2026. The limits and pricing on this page describe Claude Sonnet 5, the active successor.

Migrate to:Claude Sonnet 5Same speed class with 2× cheaper output than legacy 3.5 Sonnet pricing, 1M token context, introductory pricing ($2/$10) through August 31, 2026, and cache reads excluded from ITPM.
AnthropicGenerally available

API model string

claude-sonnet-5

Context window

1M tokens

Max output 128K tokens (sync), 300K tokens (batch)

Knowledge cutoff
not published
Released
June 30, 2026
Modalities
text + image in, text out

Last verified July 10, 2026

Rate limits by tier

Anthropic replaced its numeric Tier 1–4 system with three named tiers — Start, Build, Scale — on June 26, 2026, plus a Custom (sales) tier above Scale. Monthly spend caps are Start $500, Build $1,000, Scale $200,000. Claude Sonnet 5 shares the same RPM/ITPM/OTPM ceilings as Opus 4.x at every tier. All figures are per Anthropic docs via third-party trackers — verify actual limits in your console at platform.claude.com/settings/limits, as limits vary by organization.

TierRequirementsRPMTPMRPDConcurrentNotes
StartFirst API key or cumulative deposit under $40; no prior spend required~50 (per third-party trackers — confirm in your console/dashboard)ITPM not published separately for Sonnet 5 at Start tier; legacy Tier-1 Opus reference was ~30,000 ITPM; anecdotal post-Colossus reports cite ~500,000 ITPM (per third-party trackers — confirm in your console/dashboard)not publishednot publishedLegacy Tier 1 equivalent. Adequate for testing and prototyping only. Sonnet 5 is NOT supported on Priority Tier per Anthropic service-tiers docs (July 2026).
BuildCumulative spend above ~$40; promotion is automaticnot publishednot publishednot publishednot publishedAutomatic on crossing spend threshold; RPM and TPM scale above Start tier ceilings.
ScaleCumulative deposit threshold (~$200 equivalent); spend cap $200,000/month~4,000 (per third-party trackers — confirm in your console/dashboard)~2,000,000 ITPM (Sonnet/Opus class); ~400,000–800,000 OTPM (per third-party trackers — confirm in your console/dashboard)not publishednot publishedScale ceiling is the primary planning target for production workloads. Cache reads do NOT count toward ITPM for Sonnet 5, effectively increasing throughput headroom. Higher limits via Custom tier.
CustomContact Anthropic salesnot publishednot publishednot publishednot publishedNo spend cap; negotiated RPM/ITPM/OTPM ceilings; dedicated support SLA.

Swipe the table sideways to see every limit column.

  • 1.Token-bucket replenishment is continuous, not fixed-window reset. Anthropic acceleration limits can trigger 429 on sudden usage spikes even when you are within tier ceilings — ramp traffic gradually.
  • 2.Cache reads do NOT count toward ITPM for Sonnet 5 (current non-† classified model). This is a key throughput optimization.
  • 3.Sonnet 4.x/5 and Opus 4.x share identical tier ceilings as of June 2026 and share one org-level pool per model class.

Limits verified against the Anthropic docs, July 10, 2026.

Token pricing

What you pay per million tokens (USD). Input and output are billed separately.

Input

$2.00

per 1M tokens

Output

$10.00

per 1M tokens

Cached input

$0.20

per 1M cached

Batch discount

50%

on batch jobs

  • INTRODUCTORY pricing through August 31, 2026 only. Standard pricing flips to $3.00 input / $15.00 output per MTok on September 1, 2026. Lock in batch workloads before this date.
  • Cache read = $0.20/MTok (0.1× base input $2.00). 5-minute cache write = $2.50/MTok (1.25× base). 1-hour cache write = $4.00/MTok (2× base). Cache reads excluded from ITPM.
  • Batch API (50% off): introductory $1.00 input / $5.00 output per MTok. Async within 24 hours. Max 300K output tokens per request with output-300k-2026-03-24 beta header.
  • Extended thinking tokens billed as output (can multiply output cost 3–5×). Disable for non-reasoning tasks.
  • US-only data residency (inference_geo: 'us') adds 1.1× surcharge on Sonnet 4.6 and later. Omit unless legally required.
  • Web search tool adds $10 per 1,000 searches on top of token costs. Cache tool-calling contexts to minimize repeated searches.
  • No long-context surcharge on 1M-context Sonnet 5. No free API tier. No first-party embeddings or fine-tuning.

Side-project chatbot

$20

per month

Assumptions

5M input tokens / 1M output tokens per month, no caching

5M × $2.00/MTok = $10 input + 1M × $10.00/MTok = $10 output = $20/mo

Mid-scale app with caching

$164.40

per month

Assumptions

60M input / 12M output per month, 70% of input served from cache

18M × $2.00/MTok = $36.00 (non-cached) + 42M × $0.20/MTok = $8.40 (cache reads) + 12M × $10.00/MTok = $120.00 output = $164.40/mo

High-volume production

$1,600

per month

Assumptions

400M input / 80M output per month, no caching

400M × $2.00/MTok = $800 input + 80M × $10.00/MTok = $800 output = $1,600/mo

Run your own numbers

Drag your real monthly token volumes and watch the bill update live — priced against rival models at the same usage.

30M
1M500M
5M
0.1M100M

Estimated claude-sonnet-5 spend

$110/mo

Input: $60

Output: $50

30M in × $2.00 + 5M out × $10.00 = $110

Same volume, priced across models

  • Gemini 2.5 ProCheapest$88
  • claude-sonnet-5This model$110
  • Claude 4 Opus$275
  • GPT-5$300

Rivals priced at their published input/output rates for the same monthly volumes. Prompt caching is model-specific, so it is applied to claude-sonnet-5 only. Estimates for comparison; real bills vary with request shape and long-context surcharges.

claude-sonnet-5 vs the alternatives

Claude Sonnet 5 competes primarily with Claude Opus 4.8 (same provider, higher capability), GPT-5.5 (OpenAI), and Gemini 2.5 Pro (Google) on price, context, and throughput.

Aspectclaude-sonnet-5Claude 4 OpusGPT-5Gemini 2.5 Pro
Context window1M tokens1M tokens~1M tokens (922K input)1M+ tokens
Input $/MTok (intro)$2.00$5.00$5.00$1.25 (≤200K)
Output $/MTok (intro)$10.00$25.00$30.00$10.00
Cached input $/MTok$0.20$0.50$0.50~$0.125
Batch discount50%50%50%50%
Max output (sync)128K128K128Knot published
Thinking tokensbilled as outputbilled as outputbilled as outputbilled as output
Long-context surchargenonenone2× above 272K2× above 200K
Priority Tier supportNOT supportedyesyesn/a

Swipe the table sideways to see every model.

Hitting a 429? The playbook

The exact errors you'll see

rate_limit_errortype: "rate_limit_error"HTTP 429 Too Many Requests

Why it happens & how to fix it

RPM burst: sending requests too fast within a short window

Spread requests evenly; Anthropic's token-bucket uses acceleration limits that fire on sudden spikes even within tier ceilings. Ramp up linearly over minutes, not seconds.

ITPM saturation from large prompts or high-context inputs

Activate prompt caching with cache_control breakpoints. Cache reads do NOT count toward ITPM for Sonnet 5, effectively multiplying throughput. Use 1-hr cache writes ($4.00) for long-lived system prompts.

OTPM exceeded on high-volume streaming output

Reduce max_tokens per request; move bulk generation workloads to Batch API (50% off, async within 24h, up to 300K output tokens/request).

Org-level Sonnet pool exhausted by multiple teams sharing one key

Check the Rate Limits API (GET /v1/rate-limits, launched April 25, 2026) to see remaining headroom per dimension. Redistribute across separate API keys or teams.

Attempting to route Sonnet 5 to Priority Tier

Sonnet 5 is NOT supported on Priority Tier per Anthropic's service-tiers docs (July 2026). Use standard Start/Build/Scale tier path or upgrade to Custom for higher limits.

Retry strategy

On receiving a 429, read the retry-after header (value is seconds to wait before retrying). Also parse the anthropic-ratelimit-requests-remaining and anthropic-ratelimit-tokens-remaining headers to identify whether you are RPM-bound or TPM-bound — this determines whether you should reduce request rate or reduce token count per request. Use exponential backoff with jitter: start at 1 second, double on each retry, cap at 64 seconds.

retry.ts
1const ANTHROPIC_API_URL = "https://api.anthropic.com/v1/messages";
2
3async function callWithRetry(
4 payload: object,
5 maxRetries = 5
6): Promise<Response> {
7 let attempt = 0;
8 let delayMs = 1000;
9
10 while (attempt <= maxRetries) {
11 const res = await fetch(ANTHROPIC_API_URL, {
12 method: "POST",
13 headers: {
14 "x-api-key": process.env.ANTHROPIC_API_KEY!,
15 "anthropic-version": "2023-06-01",
16 "content-type": "application/json",
17 },
18 body: JSON.stringify(payload),
19 });
20
21 if (res.status !== 429) return res;
22
23 const retryAfter = res.headers.get("retry-after");
24 const requestsRemaining = res.headers.get("anthropic-ratelimit-requests-remaining");
25 const tokensRemaining = res.headers.get("anthropic-ratelimit-tokens-remaining");
26
27 console.warn(
28 `429 received (attempt ${attempt + 1}/${maxRetries + 1}). ` +
29 `Requests remaining: ${requestsRemaining}, Tokens remaining: ${tokensRemaining}. ` +
30 `Retry-After: ${retryAfter}s`
31 );
32
33 const waitMs = retryAfter
34 ? parseInt(retryAfter, 10) * 1000
35 : delayMs + Math.random() * 500;
36
37 await new Promise((r) => setTimeout(r, waitMs));
38 delayMs = Math.min(delayMs * 2, 64000);
39 attempt++;
40 }
41
42 throw new Error(`Exceeded max retries (${maxRetries}) on 429`);
43}

How to raise your limits

The ladder from the starter tier to enterprise — what each rung takes, and what it unlocks.

1

Start

Immediate

Automatic on creating first API key or depositing under $40 cumulative. No action required.

Unlocks: ~50 RPM; ITPM not published; spend cap $500/month. Adequate for prototyping and testing only.

2

Build

Automatic within hours to days of crossing threshold

Automatic when cumulative spend crosses ~$40 threshold. No application required.

Unlocks: RPM and ITPM ceilings scale above Start. Spend cap $1,000/month.

3

Scale

Automatic within days of crossing threshold

Automatic when cumulative deposit crosses ~$200 threshold (legacy Tier-3 equivalent). Check platform.claude.com/settings/limits for your current position.

Unlocks: ~4,000 RPM / ~2,000,000 ITPM / ~400,000–800,000 OTPM (per third-party trackers — confirm in console). Spend cap $200,000/month. Cache reads excluded from ITPM. This is the primary production planning ceiling.

4

Custom

Days to weeks to negotiate

Contact Anthropic sales team. Discuss volume commitments, use case, and enterprise agreements. Annual volume commitments can accelerate access.

Unlocks: No spend cap. Negotiated RPM/ITPM/OTPM above Scale. Dedicated support SLA. Note: Sonnet 5 Priority Tier access is NOT available at any tier — this is a per-model restriction, not a tier restriction.

Cut your token spend

Prompt caching

−90% cost on repeated context; cache reads excluded from ITPM (effective throughput multiplier)

Add cache_control breakpoints to system prompts and repeated context blocks. Use 1-hr cache writes ($4.00/MTok) for long-lived system prompts. Cache reads cost $0.20/MTok (0.1× base) AND do not consume ITPM quota — dual benefit for high-context apps.

Batch API

50% off input and output costs; removes OTPM pressure from real-time path

Route all non-latency-sensitive work to Batch API (async within 24h). At intro pricing: $1.00/$5.00 per MTok. Supports up to 300K output tokens per request with output-300k-2026-03-24 beta header.

Lock in workloads before September 1, 2026

Saves 33–50% versus post-September standard pricing ($3/$15 vs $2/$10)

Introductory pricing ($2.00 input / $10.00 output) ends August 31, 2026. Batch and cache-heavy jobs scheduled before this date lock in meaningful budget savings.

Ramp traffic gradually

Avoids acceleration-triggered 429s that fire within tier ceilings

Anthropic uses token-bucket replenishment with acceleration limits. Smooth request curves over minutes, not seconds. Never burst from 0 to max rate instantaneously.

Set conservative max_tokens

Prevents TPM exhaustion on variable-length jobs

Token estimate for a request counts toward TPM at admission time. Setting conservative max_tokens prevents OTPM exhaustion when many parallel requests are in flight.

Disable extended thinking on non-reasoning tasks

Avoids 3–5× output cost multiplier

Extended thinking tokens are billed as output tokens. Enable only when deep reasoning is genuinely needed; disable for summarization, extraction, and routing tasks.

Skip US-only data residency unless required

Saves 10% on applicable Sonnet versions

Setting inference_geo: 'us' adds 1.1× surcharge on Sonnet 4.6 and later. Omit this parameter entirely unless you have a legal or compliance requirement for US data residency.

Cache web search tool results

Reduces $10/1K searches surcharge on repeated queries

The web search tool charges $10 per 1,000 searches on top of token costs. Cache tool-calling contexts with search results so the same search is not repeated across sessions.

Frequently asked questions

How do I increase my Claude Sonnet 5 rate limits?

Rate limits increase automatically as your cumulative spend crosses thresholds: ~$40 cumulative moves you from Start to Build, and ~$200 cumulative unlocks the Scale tier (~4,000 RPM / ~2M ITPM per third-party trackers). Above Scale, contact Anthropic sales for a Custom tier with negotiated limits and no spend cap. Monitor your current tier at platform.claude.com/settings/limits or via the Rate Limits API (GET /v1/rate-limits).

Is the Claude Sonnet 5 API free?

No. There is no free tier for Claude Sonnet 5. The Start tier requires an API key and allows spend up to $500/month, but all token usage is billed. Introductory pricing is $2.00 input / $10.00 output per MTok through August 31, 2026, after which it reverts to $3.00/$15.00.

Does Claude 3.5 Sonnet still work?

Claude 3.5 Sonnet was dropped from Anthropic's active pricing record on February 19, 2026. The legacy API string remains callable, but Anthropic no longer recommends it for new projects. The current default price/performance model is Claude Sonnet 5 (claude-sonnet-5), released June 30, 2026.

What does the Claude API rate_limit_error mean?

rate_limit_error (HTTP 429) means you have exceeded one of three limits: RPM (requests per minute), ITPM (input tokens per minute), or OTPM (output tokens per minute). Check the anthropic-ratelimit-requests-remaining and anthropic-ratelimit-tokens-remaining response headers to identify which dimension you hit. The retry-after header tells you how many seconds to wait before retrying.

How does Claude Sonnet 5 pricing compare to GPT-5?

At introductory rates, Claude Sonnet 5 is $2.00/$10.00 per MTok (input/output) versus GPT-5.5 at approximately $5.00/$30.00. Sonnet 5 also has no long-context surcharge on its 1M context window, while GPT-5.5 charges 2× above 272K tokens. Both offer 50% batch discounts. Note: Sonnet 5 intro pricing ends August 31, 2026.

Do cache reads count toward Claude Sonnet 5 ITPM limits?

No — cache reads do NOT count toward ITPM for Claude Sonnet 5 (the current non-† model). This is a significant throughput advantage: heavy prompt caching both cuts cost (cache reads = $0.20/MTok vs $2.00 standard input) and preserves ITPM headroom, effectively multiplying your throughput capacity without upgrading tiers.

Can I use Claude Sonnet 5 on Priority Tier for lower latency?

No. Per Anthropic's service-tiers documentation (July 2026), Claude Sonnet 5 is NOT supported on Priority Tier. This is a per-model restriction, not a tier restriction — even Custom-tier customers cannot access Priority Tier for Sonnet 5. If Priority Tier access is a requirement, Claude Opus 4.8 supports it.

What's the best way to manage Claude API costs across a team?

Use the Rate Limits API (GET /v1/rate-limits, launched April 25, 2026) to programmatically monitor remaining headroom per dimension. Separate API keys across teams to partition the org-level pool. RapidDev's engineers help teams implement quota management, prompt caching strategies, and model routing to reduce Claude API spend — reach out at rapidevelopers.com/contact for a free scoping call.

When does Claude Sonnet 5 introductory pricing end?

Introductory pricing ($2.00 input / $10.00 output per MTok) ends August 31, 2026. Starting September 1, 2026, standard pricing takes effect at $3.00 input / $15.00 output per MTok. Batch API pricing also adjusts proportionally. Schedule your high-volume batch workloads before this date to lock in current rates.

RapidDev

We build AI apps that don't hit rate limits

  • Retry, backoff & caching built in
  • Multi-provider fallback routing
  • Fixed price, you own the code
Get a free estimate

30-min call. No commitment.

Still weighing your options?

Talk to a team that ships on all of these platforms. A free consultation gets you an honest recommendation for your specific project — even if the answer is a tool, not us.

Book a free consultation

We put the rapid in RapidDev

Need a dedicated strategic tech and growth partner? Discover what RapidDev can do for your business! Book a call with our team to schedule a free, no-obligation consultation. We'll discuss your project and provide a custom quote at no cost.