28 models on LLM.API accept the same requests as Claude Opus 4.7. They are rated 1.0–5.0 from our own catalogue data, benchmarked where independent scores exist, and priced against Claude Opus 4.7, so you can see what a swap costs before you make it.
Chosen automatically from rating, published price and capability match, updated weekly. Every point comes from published data, including where Claude Opus 4.7 is still the better choice.
$0.30 per 1M input tokens against $5.00 for Claude Opus 4.7 — 94% less.
Specs, pricing & snippet →Too expensive$0.25 per 1M input tokens against $5.00 for Claude Opus 4.7 — 95% less.
Specs, pricing & snippet →Too expensive$0.30 per 1M input tokens against $5.00 for Claude Opus 4.7 — 94% less.
Specs, pricing & snippet →Runs on Google capacity, separate from Anthropic, so a throttle on one does not stop the other.
Specs, pricing & snippet →Rate limitsRuns on Google capacity, separate from Anthropic, so a throttle on one does not stop the other.
Specs, pricing & snippet →Rate limitsRuns on DeepSeek capacity, separate from Anthropic, so a throttle on one does not stop the other.
Specs, pricing & snippet →1M tokens of context against 1M for Claude Opus 4.7.
Specs, pricing & snippet →Context too short1M tokens of context against 1M for Claude Opus 4.7.
Specs, pricing & snippet →Context too short1.1M tokens of context against 1M for Claude Opus 4.7.
Specs, pricing & snippet →Released 2026-07-21, no retirement date announced.
Specs, pricing & snippet →Deprecation or version changesReleased 2026-05-07, no retirement date announced.
Specs, pricing & snippet →Deprecation or version changesReleased 2026-09-10, no retirement date announced.
Specs, pricing & snippet →A Google model with its own safety tuning.
Specs, pricing & snippet →Too many refusalsA Google model with its own safety tuning.
Specs, pricing & snippet →Too many refusalsA DeepSeek model with its own safety tuning; open weights let you set your own policy when self-hosted.
Specs, pricing & snippet →Published weights: self-host it in your own region or private cloud, with the same model you test here.
Specs, pricing & snippet →Data privacy or residencyPublished weights: self-host it in your own region or private cloud, with the same model you test here.
Specs, pricing & snippet →Data privacy or residencyPublished weights: self-host it in your own region or private cloud, with the same model you test here.
Specs, pricing & snippet →Pick the stage closest to yours. Costs use each model's published per-token list price and assume 75% of tokens are input, which is typical for chat and RAG. Your own ratio will move the numbers, not the ranking much.
| Model | Monthly cost | Per month | Per year | vs Claude Opus 4.7 / year |
|---|---|---|---|---|
| $0.5 | $6.00 | Save $594 | ||
| $2.25 | $27.00 | Save $573 | ||
| $2.62 | $31.50 | Save $569 | ||
| $2.81 | $33.75 | Save $566 | ||
| $4.25 | $51.00 | Save $549 | ||
| $50.00 | $600 | — |
| Model | Monthly cost | Per month | Per year | vs Claude Opus 4.7 / year |
|---|---|---|---|---|
| $5.00 | $60.00 | Save $5,940 | ||
| $22.50 | $270 | Save $5,730 | ||
| $26.25 | $315 | Save $5,685 | ||
| $28.13 | $338 | Save $5,663 | ||
| $42.50 | $510 | Save $5,490 | ||
| $500 | $6,000 | — |
| Model | Monthly cost | Per month | Per year | vs Claude Opus 4.7 / year |
|---|---|---|---|---|
| $25.00 | $300 | Save $29,700 | ||
| $112 | $1,350 | Save $28,650 | ||
| $131 | $1,575 | Save $28,425 | ||
| $141 | $1,688 | Save $28,313 | ||
| $213 | $2,550 | Save $27,450 | ||
| $2,500 | $30,000 | — |
| Model | Monthly cost | Per month | Per year | vs Claude Opus 4.7 / year |
|---|---|---|---|---|
| $100.00 | $1,200 | Save $118,800 | ||
| $450 | $5,400 | Save $114,600 | ||
| $525 | $6,300 | Save $113,700 | ||
| $563 | $6,750 | Save $113,250 | ||
| $850 | $10,200 | Save $109,800 | ||
| $10,000 | $120,000 | — |
Prices per 1M tokens. Green is cheaper than Claude Opus 4.7's input price, red is dearer. Index and speed come from Artificial Analysis.
| Model | Rating | Input | Output | Context | Index | Speed | Open weights |
|---|---|---|---|---|---|---|---|
| ★ 3.6 | $5.00 | $25.00 | 1M | — | — | No | |
| ★ 4.1 | $0.30 −94% | $2.50 | 1M | 23 | 348 tok/s | No | |
| ★ 4.0 | $0.25 −95% | $1.50 | 1M | — | — | No | |
| ★ 3.9 | $0.30 −94% | $1.20 | 1M | 40 | 199 tok/s | Yes | |
| ★ 3.9 | $0.15 −97% | $0.50 | 1M | — | — | Yes | |
| ★ 3.9 | $0.50 −90% | $3.00 | 1M | — | — | No | |
| ★ 3.9 | $0.50 −90% | $3.00 | 1M | 26 | 65 tok/s | No | |
| ★ 3.8 | $0.20 −96% | $1.20 | 1.1M | 38 | 110 tok/s | No | |
| ★ 3.8 | $0.20 −96% | $1.25 | 400K | — | — | No | |
| ★ 3.8 | $0.04 −99% | $0.40 | 128K | — | — | No | |
| ★ 3.8 | $0.50 −90% | $3.00 | 1M | — | — | No | |
| ★ 3.7 | $0.60 −88% | $2.40 | 1M | 30 | 92 tok/s | Yes | |
| ★ 3.6 | $0.20 −96% | $0.50 | 262K | 40 | 53 tok/s | Yes | |
| ★ 3.6 | $0.74 −85% | $3.50 | 262K | 26 | 50 tok/s | Yes | |
| ★ 3.6 | $5.00 | $25.00 | 1M | — | — | No | |
| ★ 3.5 | $2.00 −60% | $6.00 | 500K | 44 | 55 tok/s | No | |
| ★ 3.5 | $5.00 | $25.00 | 1M | 51 | 52 tok/s | No | |
| ★ 3.5 | $2.00 −60% | $6.00 | 500K | 39 | 51 tok/s | No | |
| GLM-4.6V | ★ 3.5 | $0.30 −94% | $0.90 | 131K | — | — | Yes |
| ★ 3.5 | $0.24 −95% | $0.97 | 1M | — | — | No | |
| ★ 3.4 | $0.95 −81% | $4.00 | 262K | — | — | Yes | |
| ★ 3.4 | $0.25 −95% | $0.50 | 164K | — | — | Yes | |
| ★ 3.4 | $0.1 −98% | $0.30 | 128K | — | — | No | |
 Amazon Nova Lite | ★ 3.4 | $0.06 −99% | $0.24 | 300K | 7 | 169 tok/s | No |
| ★ 3.4 | $0.1 −98% | $0.1 | 131K | — | — | No | |
| GLM-5V-Turbo | ★ 3.3 | $1.20 −76% | $4.00 | 200K | — | — | No |
| ★ 3.3 | $0.17 −97% | $0.66 | 131K | — | — | No | |
| Amazon Nova Pro | ★ 3.0 | $0.80 −84% | $3.20 | 300K | 7 | — | No |
| Video O1 | ★ 2.6 | $15.00 +200% | $60.00 | 200K | — | — | No |
Showing the top 12 of 28. Pick a filter to see every match.
Independent scores from Artificial Analysis: the intelligence index blends reasoning, knowledge, maths and coding tests; output speed is measured tokens per second. "Input $ per index point" shows how much quality each dollar buys.
| Model | Intelligence index | Output speed | Input $ per index point |
|---|---|---|---|
| 51 | 52 tok/s | $0.098 | |
| 44 | 55 tok/s | $0.0455 | |
| 40 | 199 tok/s | $0.0075 | |
| 40 | 53 tok/s | $0.00502 | |
| 39 | 51 tok/s | $0.0513 | |
| 38 | 110 tok/s | $0.00526 | |
| 30 | 92 tok/s | $0.02 | |
| 26 | 65 tok/s | $0.0192 | |
| 26 | 50 tok/s | $0.0285 | |
| 23 | 348 tok/s | $0.013 | |
| Not published | Not published | — |
Source: Artificial Analysis. 12 of 28 alternatives have published scores; the rest are shown without one rather than estimated. Always confirm on your own prompts.
Different pipelines stress different numbers. Here is what actually drives the choice for each one, which model wins on that number, and how it stacks up against Claude Opus 4.7.
What matters: Long inputs, lots of them: context size and input price matter most.
Retrieval pipelines send far more tokens in than they get back — often 20 to 50 input tokens for every output token. That makes input price and context size the two numbers that decide your bill. A bigger window also means fewer chunks, simpler retrieval logic and less risk of cutting off the passage that holds the answer.
What matters: Needs reliable function calls and JSON output across many steps.
Agents fail in quiet ways: a malformed JSON argument or a skipped tool call breaks the loop several steps later. Native tool calling plus JSON mode is the baseline, and general reasoning strength matters because the model plans each next step itself. The intelligence index is the best public proxy for that.
Tool calling and JSON mode, rated ★ 3.5, intelligence index 51.
What matters: Data cannot leave your cloud or region.
If data residency or an air-gapped network is a hard requirement, no hosted-only model qualifies, however good it is. Open weights give you an exit: prototype through the API today, then run the same model inside your own cloud when the compliance review lands, without rewriting prompts.
1.1× Claude Opus 4.7.
What matters: Classification, tagging, summaries at volume where cost per token dominates.
Tagging, classification and summaries at volume rarely need a frontier model. Here cost per token dominates and a small quality drop is invisible after aggregation. We only consider models rated 3.0 or higher, so the cheapest pick is still a reliable one.
99% cheaper than Claude Opus 4.7.
What matters: Code generation, review and refactoring inside editors or CI.
Code work punishes small mistakes: one wrong import and the suggestion is useless. Models built or tuned for code usually beat general models of the same size, and a high intelligence index (which includes coding tests) is the next best signal.
Built or tuned for code.
Point your existing OpenAI client at LLM.API and change the model name. Nothing else in your code moves.
# Same OpenAI-compatible request — only the model name changes.
curl https://api.llmapi.ai/v1/chat/completions \
-H "Authorization: Bearer $LLMAPI_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "gemini-3.5-flash-lite",
"messages": [{"role":"user","content":"Hello!"}]
}'Every model listed here is served on LLM.API through the same OpenAI-compatible /v1/chat/completions endpoint, and each one accepts the same input types as Claude Opus 4.7. In most cases you change the "model" string and nothing else. Provider-specific parameters are listed on each model page, so check those before you switch a production prompt.
No. One LLM.API key covers every model on this page, with one invoice and one set of usage logs, so you can A/B two models against the same traffic without opening new accounts.
It comes from our own catalogue data, not paid reviews: 40% capability (price tier and how recent the model is), 35% value for money compared with similar models, and 25% context size plus support for tools, JSON and streaming. Each model has one rating, and it is the same on every page it appears on. It updates weekly.
The intelligence index and output speed are published by Artificial Analysis, an independent benchmarking site. We only show a score where they publish one; models without a score show a dash rather than an estimate.
It depends on the workload. Swap the top-rated alternative in for a sample of real traffic, compare the outputs side by side, and keep whichever wins. Because both models sit behind the same key, running that comparison costs you only the tokens.
Prices come straight from the LLM.API catalogue and refresh every Sunday. Where a provider has not published a rate, the table shows a dash instead of an estimate.
One API key for every model on this page. Run Claude Opus 4.7 and any alternative side by side on your real traffic, then keep whichever is cheaper, faster or better.
# your existing OpenAI client client = OpenAI( base_url="https://api.llmapi.ai/v1", ) client.chat.completions.create( model="claude-opus-4-7" model="gemini-3.5-flash-lite", messages=msgs, )
