Best overall
Anthropic
Model comparison
OpenRouter Models Compared
Which model should you use? Context window, price per million tokens, modality and the workload each one is actually good at — read from the public catalogue, dated, and drift-checked daily.
Best picks right now
Best for coding
Anthropic
Best for long context
Meta
Best for budget
Moondream
Best multimodal
Best for agents
Anthropic
Free and anonymous, next to what it becomes
One model below is free because its maker has not claimed it yet. The other two were free for exactly the same reason, and are now ordinary paid products. Putting them side by side is the argument for treating a stealth endpoint as an evaluation rather than a plan.
| Model | Status | Context | Input | Price per 1M | What it is |
|---|---|---|---|---|---|
| Union Alpha | Revealed 2026-09-18 | 262K | Text + Vision | $2.50 | Was the unclaimed stealth release; revealed as Unbiased and delisted on 2026-09-18 |
| Xiaomi MiMo-V2.5 | Revealed — was Hunter Alpha | 1.05M | Text + Vision | $0.119 in / $0.238 out | Named vendor, published pricing, stable endpoint |
| Z.ai GLM 5.3 Flash | Revealed — was OX Alpha | 1.05M | Text + Vision | $0.005 in / $0.199 out | Named vendor, published pricing, stable endpoint |
Source
openrouter.ai/api/v1/models Verified 2026-10-23 Union Alpha price re-read live from /api/union-alpha/status The full register — what each anonymous release became, and what it costs now — is on the stealth models index.
All models in this comparison
| Model | Vendor | Context | Input /M | Output /M | Cached in /M | Modalities |
|---|---|---|---|---|---|---|
| Claude Opus 5.5 | Anthropic | 1M | $4.00 | $20.00 | $0.200 20× cheaper
| Text, Vision, Files |
| Claude Sonnet 5.5 | Anthropic | 1M | $2.00 | $10.00 | $0.100 20× cheaper
| Text, Vision, Files |
| GPT-6.1 Sol | OpenAI | 1.05M | $2.00 | $10.00 | $0.100 20× cheaper
| Text, Vision, Files |
| Gemini 4 Argon | 1.05M | $2.00 | $10.00 | $0.100 20× cheaper
| Text, Vision, Audio, Video, Files | |
| Muse Wonder | Meta | 1.05M | $4.00 | $20.00 | $0.200 20× cheaper
| Text, Vision, Audio, Video, Files |
| Muse Spark Open | Meta | 1.05M | $1.25 | $4.25 | $0.150 8× cheaper
| Text, Vision, Audio, Video, Files |
| Moondream 3.1 | Moondream | 32K | $0.300 | $2.50 | — | Text, Vision |
| Grok 4.7 | SpaceXAI | 500K | $2.00 | $6.00 | $0.500 4× cheaper
| Text, Vision, Files |
| Gemini 3.8 Flash | 1.05M | $0.825 | $4.13 | $0.083 10× cheaper
| Text, Vision, Audio, Video, Files | |
| Nemotron 3.5 Super VL | NVIDIA | 1.05M | $0.150 | $0.500 | $0.020 8× cheaper
| Text, Vision, Video |
| Gemini Nano Banana 2.1 | 66K | $1.50 | $7.50 | — | Text, Vision | |
| Claude Haiku 5.5 | Anthropic | 1M | $0.100 | $0.500 | $0.010 10× cheaper
| Text, Vision, Files |
| Mistral Large 4 | Mistral | 1.05M | $1.36 | $4.18 | $0.140 10× cheaper
| Text, Vision |
| Command A+ | Cohere | 192K | $0.300 | $1.50 | $0.150 2× cheaper
| Text, Vision |
| Pareto 26.9 | Unbiased | 262K | $2.50 | $7.50 | $0.250 10× cheaper
| Text, Vision |
| DeepSeek V4.1 Flash | DeepSeek | 1.05M | $0.165 | $0.660 | $0.0060 28× cheaper
| Text, Vision |
| GPT-6 Astra | OpenAI | 1.05M | $10.00 | $50.00 | $1.00 10× cheaper
| Text, Vision, Files |
| Step 5 Preview | StepFun | 1.05M | $1.00 | $2.70 | $0.100 10× cheaper
| Text, Vision, Video |
| ScaleDown Extract | ScaleDown | 1M | $0.040 | $0 | — | Text, Vision, Files |
| ScaleDown Summarize | ScaleDown | 1M | $0.040 | $0 | — | Text, Vision, Files |
| Ling 3.1 Flash | inclusionAI | 262K | $0.075 | $0.225 | $0.015 5× cheaper
| Text |
| LFM2.5-24B-A2B | Liquid AI | 128K | $0.050 | $0.700 | $0.025 2× cheaper
| Text |
| Xiaomi MiMo-V2.5
formerly Hunter Alpha | Xiaomi | 1.05M | $0.119 | $0.238 | $0.0026 47× cheaper
| Text, Vision, Audio, Video |
| Xiaomi MiMo-V2.6-Flash | Xiaomi | 1.05M | $0.140 | $0.280 | $0.0028 50× cheaper
| Text, Vision, Audio, Video |
| Xiaomi MiMo-V2.6-Pro | Xiaomi | 1.05M | $0.435 | $0.870 | $0.0036 121× cheaper
| Text, Vision, Audio, Video |
| Xiaomi MiMo-V2.6-Pro-UltraSpeed | Xiaomi | 1.05M | $4.35 | $8.70 | $0.036 121× cheaper
| Text, Vision, Audio, Video |
| Z.ai GLM 5.3 Flash
formerly OX Alpha | Z.ai | 1.05M | $0.005 | $0.199 | $0.0052 1× cheaper
| Text, Vision, Video |
| DeepSeek V4 Flash | DeepSeek | 1.05M | $0.440 | $1.32 | $0.014 31× cheaper
| Text |
| DeepSeek V4 Pro | DeepSeek | 1.05M | $1.32 | $3.96 | $0.044 30× cheaper
| Text |
| Qwen3.8 Flash | Alibaba | 1M | $0.150 | $0.470 | $0.016 9× cheaper
| Text, Vision, Video |
| Qwen3.8 Max (0902) | Alibaba | 1M | $2.00 | $6.00 | $0.250 8× cheaper
| Text, Vision, Video |
| Google Gemini 3.7 Flash | 1.05M | $0.750 | $3.75 | $0.075 10× cheaper
| Text, Vision, Audio, Video, Files | |
| Claude Sonnet 5 | Anthropic | 1M | $2.00 | $10.00 | $0.200 10× cheaper
| Text, Vision, Files |
| Claude Opus 5 | Anthropic | 1M | $5.00 | $25.00 | $0.500 10× cheaper
| Text, Vision, Files |
| OpenAI GPT-5.6 Luna | OpenAI | 1.05M | $0.200 | $1.20 | $0.020 10× cheaper
| Text, Vision, Files |
| OpenAI GPT-5.6 Sol | OpenAI | 1.05M | $4.00 | $20.00 | $0.400 10× cheaper
| Text, Vision, Files |
| OpenAI GPT-5.6 Terra | OpenAI | 1.05M | $2.00 | $12.00 | $0.200 10× cheaper
| Text, Vision, Files |
| Mistral Medium 3.5 | Mistral AI | 262K | $1.50 | $7.50 | — | Text, Vision, Files |
| Meta Llama 4 Maverick | Meta | 524K | $0.188 | $0.652 | $0.050 4× cheaper
| Text, Vision |
| Cohere Command A | Cohere | 256K | $2.50 | $10.00 | — | Text |
Source
openrouter.ai/api/v1/models Verified 2026-10-23 Input/output/cached-input per 1M tokens; provider limits and surcharges can change“—” means the catalogue declares no input_cache_read for that model, not that we failed to read it The number nobody lists: cache-read pricing
Every comparison table on the web — including ours until today — lists one input price per model. But the catalogue publishes a second one: what you pay when the provider serves a prefix it has already seen. For an agent that resends the same system prompt and tool definitions on every turn, that cache price is not a discount, it is the price. Across the models we track, the gap runs from 5× to 50×.
| Model | Input /M | Cached /M | Multiplier | 100 turns, cached | the same, all fresh |
|---|---|---|---|---|---|
| Xiaomi MiMo-V2.6-Pro | $0.435 | $0.0036 | 121× | $0.018 | $2.17 |
| Xiaomi MiMo-V2.6-Pro-UltraSpeed | $4.35 | $0.036 | 121× | $0.180 | $21.75 |
| Xiaomi MiMo-V2.6-Flash | $0.140 | $0.0028 | 50× | $0.014 | $0.700 |
| Xiaomi MiMo-V2.5 | $0.119 | $0.0026 | 47× | $0.013 | $0.595 |
| DeepSeek V4 Flash | $0.440 | $0.014 | 31× | $0.070 | $2.20 |
| DeepSeek V4 Pro | $1.32 | $0.044 | 30× | $0.220 | $6.60 |
| DeepSeek V4.1 Flash | $0.165 | $0.0060 | 28× | $0.030 | $0.825 |
| Claude Opus 5.5 | $4.00 | $0.200 | 20× | $1.00 | $20.00 |
| Claude Sonnet 5.5 | $2.00 | $0.100 | 20× | $0.500 | $10.00 |
| GPT-6.1 Sol | $2.00 | $0.100 | 20× | $0.500 | $10.00 |
| Gemini 4 Argon | $2.00 | $0.100 | 20× | $0.500 | $10.00 |
| Muse Wonder | $4.00 | $0.200 | 20× | $1.00 | $20.00 |
| Claude Haiku 5.5 | $0.100 | $0.010 | 10× | $0.050 | $0.500 |
| Pareto 26.9 | $2.50 | $0.250 | 10× | $1.25 | $12.50 |
| GPT-6 Astra | $10.00 | $1.00 | 10× | $5.00 | $50.00 |
| Step 5 Preview | $1.00 | $0.100 | 10× | $0.500 | $5.00 |
| Google Gemini 3.7 Flash | $0.750 | $0.075 | 10× | $0.375 | $3.75 |
| Claude Sonnet 5 | $2.00 | $0.200 | 10× | $1.00 | $10.00 |
| Claude Opus 5 | $5.00 | $0.500 | 10× | $2.50 | $25.00 |
| OpenAI GPT-5.6 Luna | $0.200 | $0.020 | 10× | $0.100 | $1.00 |
| OpenAI GPT-5.6 Sol | $4.00 | $0.400 | 10× | $2.00 | $20.00 |
| OpenAI GPT-5.6 Terra | $2.00 | $0.200 | 10× | $1.00 | $10.00 |
| Gemini 3.8 Flash | $0.825 | $0.083 | 10× | $0.413 | $4.13 |
| Mistral Large 4 | $1.36 | $0.140 | 10× | $0.700 | $6.80 |
| Qwen3.8 Flash | $0.150 | $0.016 | 9× | $0.080 | $0.750 |
| Muse Spark Open | $1.25 | $0.150 | 8× | $0.750 | $6.25 |
| Qwen3.8 Max (0902) | $2.00 | $0.250 | 8× | $1.25 | $10.00 |
| Nemotron 3.5 Super VL | $0.150 | $0.020 | 8× | $0.100 | $0.750 |
| Ling 3.1 Flash | $0.075 | $0.015 | 5× | $0.075 | $0.375 |
| Grok 4.7 | $2.00 | $0.500 | 4× | $2.50 | $10.00 |
| Meta Llama 4 Maverick | $0.188 | $0.050 | 4× | $0.250 | $0.938 |
| Command A+ | $0.300 | $0.150 | 2× | $0.750 | $1.50 |
| LFM2.5-24B-A2B | $0.050 | $0.025 | 2× | $0.125 | $0.250 |
| Z.ai GLM 5.3 Flash | $0.005 | $0.0052 | 1× | $0.026 | $0.026 |
Source
openrouter.ai/api/v1/models → pricing.input_cache_read Verified 2026-10-23 Multiplier = fresh input price ÷ cache-read price, both read the same day“100 turns” = a 50,000-token prefix re-sent on 100 turns = 5M input tokens; it is our assumption, not a provider'sCache pricing is per provider and can change without the fresh price moving — npm run sync-models now checks it daily Where this shows up in practice: a harness that calls a decision model once per step re-sends the same instruction block every time, so it lives on the cache line rather than the one above. The field notes on Jev cover eleven projects that work exactly that way, including what each of them measured and which numbers are still the vendor's.
Best picks by workload
Best overall
- Claude Opus 5.5 Anthropic's flagship tier: multi-step changes across large codebases and long-horizon agent work
- Claude Sonnet 5.5 $2 / $10 per million with a 1M-token window — the balanced tier most product teams land on
- GPT-6.1 Sol $2 / $10 per million — the same price as the GPT-6 Sol it replaces, with a 6.1 revision
Best for coding
- Claude Opus 5.5 Anthropic's flagship tier: multi-step changes across large codebases and long-horizon agent work
- Claude Sonnet 5.5 $2 / $10 per million with a 1M-token window — the balanced tier most product teams land on
- GPT-6.1 Sol $2 / $10 per million — the same price as the GPT-6 Sol it replaces, with a 6.1 revision
Best for long context
- Muse Wonder 1,048,576-token window with a 943,718-token output ceiling — the largest output allowance in this hub
- Muse Spark Open Open weights: meta-models/Muse-Spark-Open-816B-A42B on Hugging Face, so the checkpoint can leave Meta's API
- Gemini 3.8 Flash The widest input set here: text, image, video, file and audio all go in at $0.75 / $3.75 per million
Best for budget
- Moondream 3.1 A 9B vision-language model with 2B active parameters, open weights on Hugging Face
- Nemotron 3.5 Super VL Open weights (nvidia/NVIDIA-Nemotron-3.5-Super-VL): 12B active parameters out of 120B, and a 1M-token window
- Claude Haiku 5.5 $0.10 in / $0.50 out per million — the cheapest row here from a frontier vendor
Best multimodal
- Gemini 4 Argon Google's flagship tier at $2 / $10 per million — the same price as GPT-6.1 Sol and half of Claude Opus 5.5's
- Muse Wonder 1,048,576-token window with a 943,718-token output ceiling — the largest output allowance in this hub
- Muse Spark Open Open weights: meta-models/Muse-Spark-Open-816B-A42B on Hugging Face, so the checkpoint can leave Meta's API
Best for agents
- Claude Opus 5.5 Anthropic's flagship tier: multi-step changes across large codebases and long-horizon agent work
- Claude Sonnet 5.5 $2 / $10 per million with a 1M-token window — the balanced tier most product teams land on
- GPT-6.1 Sol $2 / $10 per million — the same price as the GPT-6 Sol it replaces, with a 6.1 revision
Head-to-head comparisons
Claude Sonnet 5 vs OpenAI GPT-5.6 Sol
Claude Sonnet 5 has a lower output price, while GPT-5.6 Sol is positioned as a balanced frontier option with strong tooling.
Claude Opus 5 vs OpenAI GPT-5.6 Terra
Claude Opus 5 has lower output pricing than GPT-5.6 Terra, while Terra emphasizes strong multimodal and file reasoning.
DeepSeek V4 Flash vs Z.ai GLM 5.3 Flash
GLM 5.3 Flash adds vision and video input at $0.0052 in / $0.1985 out per million. DeepSeek V4 Flash is text-only at $0.44 in / $1.32 out. Both declare a 1,048,576-token window, so the window is not the tiebreaker.
Xiaomi MiMo-V2.5 vs DeepSeek V4 Flash
Xiaomi MiMo-V2.5 supports multimodal input at $0.119 in / $0.238 out per million. DeepSeek V4 Flash is text-only at $0.44 in / $1.32 out. On the 2026-10-23 reading MiMo is cheaper on both sides.
Xiaomi MiMo-V2.5 vs Z.ai GLM 5.3 Flash
Both declare a window of about 1.05M tokens. GLM 5.3 Flash costs $0.0052 in / $0.1985 out per million against MiMo-V2.5's $0.119 / $0.238, so GLM is the cheaper row on input by more than 20x. MiMo-V2.5 is the one that takes audio.
Qwen3.8 Flash vs Google Gemini 3.7 Flash
Qwen3.8 Flash is significantly cheaper and supports image/video, while Gemini 3.7 Flash adds audio and files with a broader multimodal surface.
Google Gemini 3.7 Flash vs Claude Sonnet 5
Gemini 3.7 Flash offers broader multimodal input at lower cost; Claude Sonnet 5 is usually stronger for code, writing and agent reliability.
DeepSeek V4 Pro vs Claude Opus 5
Claude Opus 5 is the premium reasoning option, while DeepSeek V4 Pro offers a 1M-token text workflow at a much lower price.
Meta Llama 4 Maverick vs DeepSeek V4 Flash
Both are low-cost large-context options; Llama 4 Maverick adds vision, while DeepSeek V4 Flash is text-only with lower output pricing.
OpenAI GPT-5.6 Luna vs Qwen3.8 Flash
GPT-5.6 Luna has a slight edge on context and tooling; Qwen3.8 Flash is cheaper and supports video input.
Xiaomi MiMo-V2.6-Flash vs Xiaomi MiMo-V2.6-Pro
Same release, same 1.05M-token window and the same four input modalities. Flash is $0.14 in / $0.28 out per million; Pro is $0.435 in / $0.87 out — 3.1x the price.
Xiaomi MiMo-V2.6-Flash vs DeepSeek V4 Flash
MiMo-V2.6-Flash costs $0.14 / $0.28 per million and takes image, audio and video. DeepSeek V4 Flash costs $0.44 to read and $1.32 to write, and takes text only — both around a 1.05M-token window.
Xiaomi MiMo-V2.6-Pro vs Xiaomi MiMo-V2.6-Pro-UltraSpeed
The catalogue describes both as the same 1T MiMo-V2.6-Pro checkpoint and says UltraSpeed matches Pro's quality. UltraSpeed costs exactly 10x more: $4.35 in / $8.70 out per million against $0.435 / $0.87.
Claude Sonnet 5.5 vs GPT-6.1 Sol
Identical on paper where it counts: both list $2 in / $10 out per million, a ~1M-token window, a 128K output ceiling and text, image and file input. The difference is the vendor line — Anthropic's balanced tier against OpenAI's mid-tier — and the cache price, where Claude's cached read is $0.10 against OpenAI's $0.10.
Claude Opus 5.5 vs Claude Sonnet 5.5
Exactly double the price for the same 1M-token window: Opus 5.5 lists $4 / $20 per million against Sonnet 5.5's $2 / $10, and both cap output at 128K with the same input modalities. Cached reads move the same 2x ($0.20 against $0.10).
Gemini 3.8 Flash vs Claude Haiku 5.5
Cheap-tier models pointed at different shapes of work. Gemini 3.8 Flash reads text, image, video, file and audio at $0.75 / $3.75 per million with a 1M-token window; Claude Haiku 5.5 reads text, image and files at $0.10 / $0.50 with the same 1M-token window. Haiku is 7.5x cheaper on input — and cannot see video or hear audio.
Grok 4.7 vs Claude Opus 5.5
Two flagships, one at half the output price. Grok 4.7 lists $2 / $6 per million with a 500K window and a 450K output ceiling; Claude Opus 5.5 lists $4 / $20 with a 1M window and 128K output. So Grok is cheaper on both sides and allows far longer answers, while Opus offers twice the input window.
DeepSeek V4.1 Flash vs Xiaomi MiMo-V2.6-Flash
DeepSeek V4.1 Flash lists $0.165 / $0.66 per million with a 943,718-token output ceiling and image input; MiMo V2.6 Flash lists $0.14 / $0.28 with image, audio and video input. MiMo is about 15% cheaper on input and 58% cheaper on output; DeepSeek allows output about seven times longer.
Muse Wonder vs Claude Opus 5.5
The same price, a different number. Muse Wonder lists $4 in / $20 out per million and so does Claude Opus 5.5, with cached reads at $0.20 for both; Artificial Analysis puts Opus 5.5 at 58 on its Intelligence Index and Muse Wonder at 52, at $5.98 and $4.52 per evaluation task. Meta's own launch table prints a different comparison and a different winner.
Gemini 4 Argon vs GPT-6.1 Sol
Two flagships at exactly the same price. Both list $2 in / $10 out per million with a ~1M-token window and cached reads at $0.10; Artificial Analysis puts Gemini 4 Argon at 53 and, on its own page, GPT-6.1 Sol's sibling GPT-6 Astra at 53 as well. Google reads video, audio, images and files; OpenAI reads images and files.
Frequently asked questions
What is the best model on OpenRouter for coding?
Claude Opus 5.5 and Claude Sonnet 5.5 lead this snapshot for coding, with GPT-6.1 Sol at the same $2 / $10 price as Sonnet and Grok 4.7 undercutting all three on output at $6 per million. Sonnet is the usual price/quality default; Opus is for multi-step changes where a wrong edit is the expensive part.
Which OpenRouter model has the largest context window?
In the 2026-10-23 snapshot, GPT-6.1 Sol offers the largest window at 1,050,000 tokens. Several others sit just behind it around 1M tokens — including free and budget rows such as Z.ai GLM 5.3 Flash. Always verify provider-specific limits before long-context production use.
What is the cheapest model on OpenRouter?
Two answers, because they are different questions. The lowest listed price in this snapshot is Z.ai GLM 5.3 Flash at $0.0052 per million input tokens. The cheapest paid row is Z.ai GLM 5.3 Flash at $0.0052. For :free routes see our free models page — those carry rate limits and smaller windows.
Are there free models on OpenRouter?
Yes. OpenRouter exposes some models with a :free suffix. These are useful for testing and low-volume workloads, but rate limits, context limits and quality vary significantly.
What happened to Hunter Alpha and OX Alpha?
Hunter Alpha is now known as Xiaomi MiMo-V2.5 and OX Alpha was revealed as GLM 5.3 Flash. We keep those names as historical aliases, but no longer treat them as primary SEO targets.
How do OpenRouter models compare to ChatGPT?
OpenRouter is not a single model; it is a model router and marketplace. The best comparison depends on the model you choose, the input modality, context length and pricing tier.