Skip to main content
RapidDev - Software Development Agency
AI API Limits & Performance Matrix11 min readVerified July 10, 2026

Yi-Large API Rate Limits, Pricing & Performance (July 2026)

01.AI's international first-party API is closed — the platform.01.ai shutdown banner has been displayed since at least August 25, 2024 (year not definitively confirmed from primary source). Fireworks AI is the sole active third-party host, serving Yi-Large at $3.00 in / $3.00 out per million tokens with a 32K–33K token context window. For new projects, Qwen3-Max or Llama 4 Scout offer larger context and lower input costs.

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

Deprecated model

Sunsets First-party: assessed August 25, 2024 (year not definitively confirmed from primary source — verify). Fireworks host: no sunset date announced as of July 10, 2026.

01.AI's international first-party API is closed. The platform.01.ai shutdown banner has been displayed since at least August 25, 2024 (year not definitively confirmed from a primary source — flagged). 01.AI pivoted away from LLM pre-training in March 2025, focusing on enterprise and DeepSeek-based solutions; the international API was not reinstated. Yi-Large is accessible only via Fireworks AI ($3/$3 per MTok) and potentially open-weight self-host. New projects should not be built on Yi-Large.

Migrate to:Qwen3-MaxActive first-party API on DashScope, 262K context, 50% batch discount, and significantly more competitive pricing at $1.20/$6.00 vs Yi-Large's $3.00/$3.00 flat rate
01.AI (first-party closed; hosted by Fireworks AI)Generally available

API model string

accounts/fireworks/models/yi-large

Context window

32K–33K tokens

Max output not published

Knowledge cutoff
not published
Released
June 2024
Modalities
text in, text out

Last verified July 10, 2026

Rate limits by tier

Yi-Large has no active first-party API — 01.AI's international platform is closed. Fireworks AI is the sole confirmed active third-party host as of July 10, 2026; all rate limits below refer to Fireworks standard policies, which are not individually published for this model.

TierRequirementsRPMTPMRPDNotes
01.AI First-Party International APICLOSED — not availablen/an/an/aplatform.01.ai shows shutdown banner: "API services will be temporarily unavailable starting from August 25th due to business adjustments." The platform redirects to the Chinese product (platform.lingyiwanwu.com / 零一万物). International API closed, assessed August 25, 2024 — year not definitively confirmed from primary source (verify).
Fireworks AI (sole active third-party host)Fireworks account; payment method for pay-as-you-gonot publishednot publishednot publishedYi-Large available at $3.00 in / $3.00 out per MTok with 32.8K–33K context. Fireworks standard rate limits apply; exact RPM/TPM not published for this specific model. Verify current availability and limits in the Fireworks console.
OpenRouterOpenRouter accountnot publishednot publishednot publishedYi-Large model page exists on OpenRouter (released Jun 25, 2024, 33K context listed) but shows no active provider or price as of July 10, 2026. Do not use as a primary path without verifying current availability.
Self-host (open weights)GPU hardware; Yi-34B or Yi-6B open weights (Apache 2.0 on Hugging Face)self-managedself-managedself-managedYi-34B and Yi-6B are released under Apache 2.0 and commercially usable. Yi-Large specific full-model weights availability on Hugging Face is not fully confirmed — verify on huggingface.co/01-ai before planning a self-host path. No caching or batch API applicable.

Swipe the table sideways to see every limit column.

  • 1.01.AI pivoted away from LLM pre-training in March 2025, focusing on enterprise/DeepSeek-based solutions; the international API was not reinstated.
  • 2.Last 01.AI model release listed November 29, 2024.
  • 3.No caching or batch API from Fireworks for Yi-Large specifically — verify on Fireworks pricing page before assuming these features are available.

Limits verified against the 01.AI (first-party closed; hosted by Fireworks AI) docs, July 10, 2026.

Token pricing

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

Input

$3.00

per 1M tokens

Output

$3.00

per 1M tokens

  • Pricing is for Fireworks AI — the sole active host. First-party 01.AI pricing is not published (API closed).
  • Yi-Large uses an unusually symmetric flat rate: same $3.00/MTok for both input and output. Verify current rate on the Fireworks AI pricing page; sourced from CloudPrice.net as of July 10, 2026.
  • No context caching or batch discount confirmed for Yi-Large on Fireworks — the 70%-cached scenario is not applicable.
  • No free tier or trial credits specific to Yi-Large; Fireworks general trial credits may apply — confirm on sign-up.

Side-project chatbot

$18.00

per month

Assumptions

5M input / 1M output tokens per month via Fireworks at $3.00/$3.00

5 × $3.00 + 1 × $3.00 = $15.00 + $3.00 = $18.00

Mid-scale app

$216.00

per month

Assumptions

60M input / 12M output per month; no caching available

60 × $3.00 + 12 × $3.00 = $180.00 + $36.00 = $216.00 (no cache discount — not available for Yi-Large on Fireworks)

High-volume pipeline

$1,440.00

per month

Assumptions

400M input / 80M output per month via Fireworks at $3.00/$3.00

400 × $3.00 + 80 × $3.00 = $1,200.00 + $240.00 = $1,440.00

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 accounts/fireworks/models/yi-large spend

$105/mo

Input: $90

Output: $15

30M in × $3.00 + 5M out × $3.00 = $105

Same volume, priced across models

  • Llama 4 ScoutCheapest$3.90
  • Kimi K2$49
  • Qwen3-Max$66
  • accounts/fireworks/models/yi-largeThis model$105

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

accounts/fireworks/models/yi-large vs the alternatives

Yi-Large on Fireworks AI compared to live alternatives with first-party APIs — all data verified July 10, 2026.

Aspectaccounts/fireworks/models/yi-largeQwen3-MaxLlama 4 ScoutKimi K2
First-party APIclosed (01.AI international)live (DashScope)third-party hostslegacy migrated to K2.6
Context window32K–33K tokens262K tokensup to 1M tokens (hosted)262K tokens
Input $/MTok$3.00 (Fireworks)$1.20 (DashScope)$0.08 (DeepInfra)$0.95 (K2.6)
Output $/MTok$3.00 (Fireworks)$6.00$0.30$4.00
Multilingual supportEN / ZH / ES / JA / DE / FRmultilingualmultilingualmultilingual
License (open weights)Apache 2.0 (Yi-6B / Yi-34B)Qwen3 licenseLlama 4 CommunityModified MIT
Active first-party developmenthalted (01.AI pivot March 2025)activeactiveactive
Host diversity1 active host (Fireworks only)DashScope + aggregatorsmany inference hostsFireworks + others

Swipe the table sideways to see every model.

Hitting a 429? The playbook

The exact errors you'll see

HTTP 429 Too Many Requests (Fireworks AI)Fireworks standard rate limit exceeded (exact error string not published — verify in Fireworks documentation)401 Unauthorized (when attempting to use the closed 01.AI first-party endpoint)

Why it happens & how to fix it

Calling the closed 01.AI first-party endpoint (api.01.ai or platform.01.ai)

The 01.AI international API is closed. Switch your base URL to Fireworks: https://api.fireworks.ai/inference/v1 with model accounts/fireworks/models/yi-large.

Fireworks RPM or TPM quota exceeded

Implement exponential backoff with jitter; honor any Retry-After header. Check your Fireworks console for current quota. Contact Fireworks support for quota increases.

Input exceeds Yi-Large's 32K–33K token context window

Truncate or summarize input before sending. Yi-Large has one of the shortest context windows in its class — plan for this hard limit. Consider Qwen3-Max (262K) for long-context use cases.

Fireworks account credits depleted

Add credits or a payment method to your Fireworks account. For high-volume workloads, evaluate migrating to Llama 4 Scout ($0.08/$0.30) to reduce costs significantly.

Retry strategy

For Fireworks AI: use exponential backoff with jitter on HTTP 429 responses. Honor the Retry-After header when present. Fireworks is OpenAI-SDK compatible — use the standard retry pattern with base_url override. Ramp request rate gradually after backoff to avoid burst re-triggering rate limits. Start with a 1-second base delay, doubling each attempt with a random jitter of 0–500ms, capped at 60 seconds.

retry.ts
1import OpenAI from "openai";
2
3const client = new OpenAI({
4 apiKey: process.env.FIREWORKS_API_KEY,
5 baseURL: "https://api.fireworks.ai/inference/v1",
6});
7
8async function callYiLarge(
9 messages: OpenAI.Chat.ChatCompletionMessageParam[],
10 maxRetries = 5
11): Promise<string> {
12 for (let attempt = 0; attempt < maxRetries; attempt++) {
13 try {
14 const response = await client.chat.completions.create({
15 model: "accounts/fireworks/models/yi-large",
16 messages,
17 max_tokens: 1024,
18 });
19 return response.choices[0].message.content ?? "";
20 } catch (err: unknown) {
21 const error = err as { status?: number; headers?: Record<string, string> };
22 if (error?.status === 429) {
23 const retryAfter = parseInt(error.headers?.["retry-after"] ?? "0", 10);
24 const base = retryAfter > 0 ? retryAfter * 1000 : Math.min(60000, 1000 * 2 ** attempt);
25 const jitter = Math.random() * 500;
26 await new Promise((r) => setTimeout(r, base + jitter));
27 continue;
28 }
29 throw err;
30 }
31 }
32 throw new Error("Max retries exceeded for Yi-Large on Fireworks");
33}

How to raise your limits

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

1

Fireworks trial

Immediate

Sign up at fireworks.ai — trial credits may be available on new accounts

Unlocks: Limited free inference for evaluation; verify current trial credit offer on Fireworks sign-up

2

Fireworks pay-as-you-go

Immediate

Add a payment method on fireworks.ai; Yi-Large is available at $3.00/$3.00 per MTok

Unlocks: Full Yi-Large access; Fireworks standard rate limits (exact RPM/TPM not published — shown in Fireworks console)

3

Fireworks enterprise

Sales cycle (days to weeks)

Contact Fireworks sales at fireworks.ai

Unlocks: Higher rate limits, SLA guarantees, priority support

4

Migrate to a live alternative (recommended for new projects)

Immediate

Switch model string to qwen3-max on DashScope intl, or llama-4-scout on Fireworks/Groq — no code changes beyond base_url and model name

Unlocks: Active first-party development, larger context (262K or 1M), lower input cost ($0.08–$1.20 vs $3.00)

Cut your token spend

Migrate to Qwen3-Max or Llama 4 Scout for new projects

2.5–37× lower input token cost

Yi-Large on Fireworks charges $3.00/MTok input — Llama 4 Scout is $0.08/MTok (37× cheaper) and Qwen3-Max is $1.20/MTok (2.5× cheaper). Only keep Yi-Large if existing integrations depend on its specific behavior.

Plan around the 32K–33K context hard limit

Prevents silent truncation errors in production

Yi-Large has one of the shortest context windows in the 2024 model generation. For conversations, RAG pipelines, or document tasks that can exceed 32K tokens, migrate to Qwen3-Max (262K) or Llama 4 Scout (up to 1M) before hitting production.

Monitor Fireworks for Yi-Large availability

Risk mitigation — sole active host

Fireworks is currently the only active third-party host for Yi-Large. If Fireworks deprioritizes or removes the model, there is no confirmed fallback. Set up alerts on your API response codes and maintain a migration plan to an alternative model.

Use Yi-Large only for existing tuned integrations

Avoids unnecessary migration risk if already deployed

If your application has been fine-tuned or prompt-engineered specifically for Yi-Large's response style, staying on Fireworks for now is lower risk than re-tuning. Set a migration milestone for Q3 2026.

For multilingual CJK workloads, evaluate Qwen3-Max

Potentially better ZH/JA language quality from Alibaba's training data

Qwen3-Max is trained by the same provider ecosystem (China-based) with deep CJK corpus coverage. If Yi-Large was selected for Chinese or Japanese language quality, Qwen3-Max is a natural migration target with active support.

Frequently asked questions

Does Yi-Large have an API?

01.AI's international first-party API at platform.01.ai is closed as of at least August 25, 2024 (year not definitively confirmed from a primary source). Fireworks AI is the sole confirmed active third-party host as of July 10, 2026, serving Yi-Large at $3.00 in / $3.00 out per million tokens. There is no official 01.AI endpoint you can sign up for directly.

How do I access Yi-Large now that the 01.AI API is closed?

Use Fireworks AI: set your base URL to https://api.fireworks.ai/inference/v1 and model to accounts/fireworks/models/yi-large. The Fireworks API is OpenAI-SDK compatible, so you only need to change the base_url and model name in your existing code. Sign up at fireworks.ai and add a payment method for pay-as-you-go access.

What are Yi-Large's rate limits?

Yi-Large's rate limits on Fireworks are not individually published — they follow Fireworks standard limits for pay-as-you-go accounts. Exact RPM and TPM are shown in your Fireworks console after signing up. The first-party 01.AI rate limits are not applicable as that API is closed.

How much does Yi-Large cost per million tokens?

$3.00 per million input tokens and $3.00 per million output tokens on Fireworks AI — an unusually symmetric flat rate. There is no published caching or batch discount for Yi-Large on Fireworks. Verify the current rate on the Fireworks pricing page, as it may change without notice.

Is Yi-Large free to use?

There is no permanent free tier for Yi-Large. Fireworks AI may offer trial credits for new accounts — check at sign-up. The former 01.AI free tier is not available as the international platform is closed. For free inference at scale, self-hosting Yi-34B or Yi-6B under the Apache 2.0 license is possible, though Yi-Large specific weights availability should be verified on Hugging Face.

Yi-Large vs Qwen3-Max: which should I use?

For new projects: Qwen3-Max. It offers 262K context (vs Yi-Large's 32K–33K), an active first-party API on DashScope at $1.20/$6.00 (vs Yi-Large's $3.00/$3.00 flat on Fireworks), and a 50% batch discount. Yi-Large's only comparative advantage is its $3.00/MTok output rate, which is lower than Qwen3-Max's $6.00 output rate — but the context and development activity gap is decisive for most use cases.

How do I increase Yi-Large rate limits on Fireworks?

Contact Fireworks sales for enterprise rate limit increases. For standard pay-as-you-go accounts, limits are set by Fireworks policy and shown in your console. If Yi-Large's limits are insufficient for your scale, RapidDev can help you evaluate multi-provider quota strategies and migration paths — reach out at rapidevelopers.com/contact.

What should I do if I get a 429 error from Yi-Large on Fireworks?

Implement exponential backoff: start with a 1-second delay, double each retry, add random jitter, and cap at 60 seconds. Honor the Retry-After header if Fireworks includes one. If 429s are frequent, check your Fireworks console for your quota ceiling and consider contacting Fireworks for an increase, or evaluate migrating to Llama 4 Scout ($0.08/$0.30) or Qwen3-Max ($1.20/$6.00) for more cost-effective high-volume inference.

Can I self-host Yi-Large?

Yi-34B and Yi-6B are available on Hugging Face under the Apache 2.0 license and can be self-hosted commercially. Yi-Large's specific full-model weights may not be fully publicly available — verify on huggingface.co/01-ai before planning a self-host deployment. For the open-weight variants, you need sufficient GPU memory: Yi-34B requires approximately 70GB at FP16 (2× A100 40GB).

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.