AGPL-3.0-or-later · forever.

the models · 64

Every model, one room.

The union of the wire catalog: each model names the providers that serve it, the window it reads, the exact-match price rows and the measured energy rows. What the catalog does not carry, the room says plainly · a missing number is a fact, never a zero.

Vendored from the released binary · supernovae-st/nika@25f502adb7ec (v0.111.0) · digest-verified, re-derived at build, gated in CI

The bench

1 published receipt: a room shows a measured answer only when a receipt names its model, with the evidence pack served beside it. Every other seat: run it yourself · the model-bench workflow in the public registry measures YOUR machine, and the first receipt shows what an honest one looks like.

The context axis

The axis draws the canonical catalog · the spec-named seats whose models declare a window. The register below is the wider wire union; each count names its facet.

29 models the catalog knows, from 32k to 1049k tokens of context. 7 of them want no key at all.

  • mistral-small-latest32k
  • deepseek-chat64k
  • qwen3.5:4b128k
  • llama3.2:3b128k
  • qwen3.5-4b128k
  • llamacpp/default128k
  • localai/default128k
  • Qwen/Qwen3-8B128k
  • mistral-large-latest128k
  • llama-3.3-70b-versatile128k
  • llama-3.1-8b-instant128k
  • anthropic/claude-sonnet-4-6128k
  • anthropic/claude-haiku-4-5-20251001128k
  • meta/llama-3.1-70b-instruct128k
  • kimi-k2128k
  • grok-3131k
  • grok-3-mini-fast131k
  • claude-sonnet-4-6200k
  • claude-haiku-4-5-20251001200k
  • kimi-k2.5256k
  • Qwen/Qwen3.5-397B-A17B:fastest262k
  • Qwen/Qwen3.5-9B:cheapest262k
  • nvidia/nemotron-3-super-120b-a12b262k
  • nvidia/nemotron-3-nano-30b-a3b262k
  • gpt-5.2272k
  • gpt-5-mini272k
  • mock-default1000k
  • gemini-2.5-flash1049k
  • gemini-2.0-flash1049k

no key needed · a key, held by you · declared reasoning · takes images. A log scale, because the range is five doublings wide. Context and capabilities derive from the catalog the engine ships; the pricing rows it carries are empty today, so this page does not draw a price it does not have.

The price axis

Every priced model, seated at its cheapest recorded output seat. The taller accent ticks are open weights: the seats you could also host yourself.

12 priced · $0.08 → $15 out/Mtok · log scale · 6 open weights

The register

  1. cloudflare · 128k context

  2. cloudflare · 128k context

  3. sambanova · 128k context

  4. sambanova · 128k context

  5. MiniMax-M21 seat · 245k

    minimax · 245k context

  6. MiniMax-VL-011 seat · 100k

    minimax · 100k context

  7. Qwen/Qwen3-8B2 seats · 128k

    native · vllm · 128k context

  8. huggingface · 262k context

  9. huggingface · 262k context

  10. fireworks · 128k context

  11. fireworks · 128k context

  12. bedrock · 128k context

  13. bedrock · 128k context

  14. openrouter · 128k context

  15. openrouter · 128k context

  16. claude-haiku-4-5-202510011 seat · 200k · $5 out

    anthropic · 200k context

  17. claude-sonnet-4-61 seat · 200k · $15 out

    anthropic · 200k context

  18. command-r1 seat · 128k

    cohere · 128k context

  19. command-r-plus1 seat · 128k

    cohere · 128k context

  20. databricks · 128k context

  21. deepseek-chat1 seat · 64k · $0.28 out⬡ open

    deepseek · 64k context

  22. default2 seats · 128k

    llamacpp · localai · 128k context

  23. gemini-2.0-flash2 seats · 1.0m

    gemini · vertex · 1.0m context

  24. gemini-2.5-flash2 seats · 1.0m

    gemini · vertex · 1.0m context

  25. glm-4-flash1 seat · 128k

    zhipu · 128k context

  26. glm-4.51 seat · 128k

    zhipu · 128k context

  27. gpt-4o1 seat · 128k

    azure · 128k context

  28. gpt-4o-mini1 seat · 128k

    azure · 128k context

  29. gpt-5-mini1 seat · 272k · $2 out

    openai · 272k context

  30. gpt-5.21 seat · 272k · $14 out

    openai · 272k context

  31. grok-31 seat · 131k · $15 out

    xai · 131k context

  32. grok-3-mini-fast1 seat · 131k · $4 out

    xai · 131k context

  33. jamba-1.5-large1 seat · 256k

    ai21 · 256k context

  34. jamba-1.5-mini1 seat · 256k

    ai21 · 256k context

  35. kimi-k21 seat · 128k

    moonshot · 128k context

  36. kimi-k2.51 seat · 256k

    moonshot · 256k context

  37. llama-3.1-8b1 seat · 128k

    cerebras · 128k context

  38. llama-3.1-8b-instant1 seat · 128k · $0.08 out⬡ open

    groq · 128k context

  39. llama-3.3-70b1 seat · 128k

    cerebras · 128k context

  40. llama-3.3-70b-versatile1 seat · 128k · $0.79 out⬡ open

    groq · 128k context

  41. llama3.2:3b1 seat · 128k

    ollama · 128k context

  42. together · 128k context

  43. hyperbolic · 128k context

  44. together · 128k context

  45. deepinfra · 1.0m context

  46. deepinfra · 128k context

  47. meta/llama-3.1-70b-instruct1 seat · 128k⬡ open

    nvidia · 128k context

  48. replicate · 8k context

  49. replicate · 8k context

  50. mistral-large-latest1 seat · 128k · $1.5 out⬡ open

    mistral · 128k context

  51. mistral-small-latest1 seat · 32k · $0.6 out⬡ open

    mistral · 32k context

  52. mock-default1 seat · 1m

    mock · 1m context

  53. nvidia/nemotron-3-nano-30b-a3b1 seat · 262k⬡ open

    nvidia · 262k context

  54. nvidia/nemotron-3-super-120b-a12b1 seat · 262k · $0.8 out⬡ open

    nvidia · 262k context

  55. palmyra-x-0041 seat · 128k

    writer · 128k context

  56. qwen-max1 seat · 32k

    qwen · 32k context

  57. qwen-turbo1 seat · 1m

    qwen · 1m context

  58. qwen-vl-plus1 seat · 32k

    qwen · 32k context

  59. qwen3.5-4b1 seat · 128k

    lmstudio · 128k context

  60. qwen3.5:4b1 seat · 128k

    ollama · 128k context

  61. sonar1 seat · 128k

    perplexity · 128k context

  62. sonar-pro1 seat · 200k

    perplexity · 200k context

  63. voyage-3-large1 seat · 32k

    voyage · 32k context

  64. voyage-3-lite1 seat · 32k

    voyage · 32k context