SPB Git forge

spb/polyllm

Public
15commits 1branches 0releases
2.2 MBsize
maindefault branch
13 days agolast push
TypeScript 97.4% SQL 1% JavaScript 0.9% CSS 0.6%
12.5 KB · 357 lines json
Raw Blame History
1[2  {3    "id": "grok-4.6",4    "displayName": "Grok 4.6",5    "family": "grok-4.6",6    "contextTokens": 500000,7    "maxOutputTokens": 128000,8    "capabilities": {9      "text": true,10      "vision": true,11      "audioInput": false,12      "audioOutput": false,13      "imageGeneration": false,14      "video": false,15      "reasoning": true,16      "tools": true,17      "structuredOutput": true,18      "streaming": true,19      "files": true,20      "webSearch": true21    },22    "parameters": {23      "temperature": true,24      "topP": true,25      "topK": "accepted-on-chat-completions-undocumented",26      "maxTokens": true,27      "reasoningEffort": ["low", "medium", "high", "xhigh"],28      "thinkingBudget": false,29      "stop": false,30      "seed": true,31      "frequencyPenalty": false,32      "presencePenalty": false33    },34    "pricing": {35      "inputPerMillion": 2.0,36      "cachedInputPerMillion": 0.5,37      "outputPerMillion": 6.0,38      "longContextThresholdTokens": 200000,39      "longContextInputPerMillion": 4.0,40      "longContextCachedInputPerMillion": 1.0,41      "longContextOutputPerMillion": 12.0,42      "imageInputPerMillionTokens": 2.043    },44    "status": "active",45    "aliases": [],46    "notes": "Flagship (Aug 2026). Always-reasoning; reasoning_effort default high, 'none' rejected ('This model does not support `reasoning_effort` value `none`'). reasoning_content returned in chat completions (plain text, also streamed as delta.reasoning_content). Knowledge cutoff 2026-02-01. Docs: 'no text output limit'; API default cap max_completion_tokens=128000. Large hidden system prompt (~640 prompt tokens on an empty request). Supports web_search/x_search/code_interpreter/mcp/file tools on /v1/responses. Rate-limit headers observed: 7200 req / 50M tokens per window."47  },48  {49    "id": "grok-4.5",50    "displayName": "Grok 4.5",51    "family": "grok-4.5",52    "contextTokens": 500000,53    "maxOutputTokens": 128000,54    "capabilities": {55      "text": true,56      "vision": true,57      "audioInput": false,58      "audioOutput": false,59      "imageGeneration": false,60      "video": false,61      "reasoning": true,62      "tools": true,63      "structuredOutput": true,64      "streaming": true,65      "files": true,66      "webSearch": true67    },68    "parameters": {69      "temperature": true,70      "topP": true,71      "topK": "accepted-on-chat-completions-undocumented",72      "maxTokens": true,73      "reasoningEffort": ["low", "medium", "high"],74      "thinkingBudget": false,75      "stop": false,76      "seed": true,77      "frequencyPenalty": false,78      "presencePenalty": false79    },80    "pricing": {81      "inputPerMillion": 2.0,82      "cachedInputPerMillion": 0.3,83      "outputPerMillion": 6.0,84      "longContextThresholdTokens": 200000,85      "longContextInputPerMillion": 4.0,86      "longContextCachedInputPerMillion": 0.6,87      "longContextOutputPerMillion": 12.0,88      "imageInputPerMillionTokens": 2.089    },90    "status": "active",91    "aliases": ["grok-4.5-latest", "grok-build-latest"],92    "notes": "Coding/agentic model (Jul 2026). Always-reasoning; 'none' rejected; 'xhigh' accepted by the API but docs say it is treated as 'high' on models that do not support it. Very few reasoning tokens on trivial prompts (~13-50). reasoning_content returned. Note the alias 'grok-build-latest' points here, not to grok-build-0.1."93  },94  {95    "id": "grok-4.3",96    "displayName": "Grok 4.3",97    "family": "grok-4.3",98    "contextTokens": 1000000,99    "maxOutputTokens": 128000,100    "capabilities": {101      "text": true,102      "vision": true,103      "audioInput": false,104      "audioOutput": false,105      "imageGeneration": false,106      "video": false,107      "reasoning": true,108      "tools": true,109      "structuredOutput": true,110      "streaming": true,111      "files": true,112      "webSearch": true113    },114    "parameters": {115      "temperature": true,116      "topP": true,117      "topK": "accepted-on-chat-completions-undocumented",118      "maxTokens": true,119      "reasoningEffort": ["none", "low", "medium", "high"],120      "thinkingBudget": false,121      "stop": false,122      "seed": true,123      "frequencyPenalty": false,124      "presencePenalty": false125    },126    "pricing": {127      "inputPerMillion": 1.25,128      "cachedInputPerMillion": 0.2,129      "outputPerMillion": 2.5,130      "longContextThresholdTokens": 200000,131      "longContextInputPerMillion": 2.5,132      "longContextCachedInputPerMillion": 0.4,133      "longContextOutputPerMillion": 5.0,134      "imageInputPerMillionTokens": 1.25135    },136    "status": "active",137    "aliases": ["grok-4.3-latest"],138    "notes": "Best value general model; 1M context. The ONLY probed model that accepts reasoning_effort 'none' (reasoning_tokens=0 -> effectively a non-reasoning mode). 'xhigh' accepted (docs/AI SDK say not supported -> treated as high). Retired May-15-2026 slugs (grok-3, grok-4-0709, grok-4-fast-*, grok-4-1-fast-*) are transparently served by grok-4.3 (response.model = 'grok-4.3'). Retired non-reasoning slugs map to effort 'none', reasoning slugs to 'low'. reasoning_content returned. Batch API 20% discount."139  },140  {141    "id": "grok-4.20-0309-reasoning",142    "displayName": "Grok 4.20 Reasoning (0309)",143    "family": "grok-4.20",144    "contextTokens": 1000000,145    "maxOutputTokens": 128000,146    "capabilities": {147      "text": true,148      "vision": true,149      "audioInput": false,150      "audioOutput": false,151      "imageGeneration": false,152      "video": false,153      "reasoning": true,154      "tools": true,155      "structuredOutput": true,156      "streaming": true,157      "files": true,158      "webSearch": true159    },160    "parameters": {161      "temperature": true,162      "topP": true,163      "topK": "accepted-on-chat-completions-undocumented",164      "maxTokens": true,165      "reasoningEffort": false,166      "thinkingBudget": false,167      "stop": false,168      "seed": true,169      "frequencyPenalty": false,170      "presencePenalty": false171    },172    "pricing": {173      "inputPerMillion": 1.25,174      "cachedInputPerMillion": 0.2,175      "outputPerMillion": 2.5,176      "longContextThresholdTokens": 200000,177      "longContextInputPerMillion": 2.5,178      "longContextCachedInputPerMillion": 0.4,179      "longContextOutputPerMillion": 5.0,180      "imageInputPerMillionTokens": 1.25181    },182    "status": "active",183    "aliases": [184      "grok-4.20-reasoning-latest",185      "grok-4.20",186      "grok-4.20-reasoning",187      "grok-4.20-0309",188      "grok-4.20-beta-0309-reasoning",189      "grok-4.20-beta",190      "grok-4.20-beta-0309",191      "grok-4.20-beta-latest",192      "grok-4.20-beta-latest-reasoning",193      "grok-4.20-beta-reasoning",194      "grok-4.20-experimental-beta-0304-reasoning",195      "grok-4.20-experimental-beta-0304",196      "grok-4.20-experimental-beta-reasoning-latest",197      "grok-4.20-experimental-beta-latest",198      "grok-4.20-reasoning-gv2"199    ],200    "notes": "Always-reasoning, fixed depth: ANY reasoning_effort value (incl. 'none') is rejected with 400 'Model grok-4.20-0309-reasoning does not support parameter reasoningEffort.' (both chat completions and /v1/responses reasoning.effort). reasoning_content returned. Occasionally emits stray '\\\\confidence{NN}' tokens after the answer on trivial prompts (observed with top_k/logprobs). Batch API 20% discount."201  },202  {203    "id": "grok-4.20-0309-non-reasoning",204    "displayName": "Grok 4.20 Non-Reasoning (0309)",205    "family": "grok-4.20",206    "contextTokens": 1000000,207    "maxOutputTokens": 128000,208    "capabilities": {209      "text": true,210      "vision": true,211      "audioInput": false,212      "audioOutput": false,213      "imageGeneration": false,214      "video": false,215      "reasoning": false,216      "tools": true,217      "structuredOutput": true,218      "streaming": true,219      "files": true,220      "webSearch": true221    },222    "parameters": {223      "temperature": true,224      "topP": true,225      "topK": "accepted-on-chat-completions-undocumented",226      "maxTokens": true,227      "reasoningEffort": false,228      "thinkingBudget": false,229      "stop": true,230      "seed": true,231      "frequencyPenalty": false,232      "presencePenalty": false233    },234    "pricing": {235      "inputPerMillion": 1.25,236      "cachedInputPerMillion": 0.2,237      "outputPerMillion": 2.5,238      "longContextThresholdTokens": 200000,239      "longContextInputPerMillion": 2.5,240      "longContextCachedInputPerMillion": 0.4,241      "longContextOutputPerMillion": 5.0,242      "imageInputPerMillionTokens": 1.25243    },244    "status": "active",245    "aliases": [246      "grok-4.20-non-reasoning",247      "grok-4.20-non-reasoning-latest",248      "grok-4.20-beta-non-reasoning",249      "grok-4.20-beta-latest-non-reasoning",250      "grok-4.20-experimental-beta-0304-non-reasoning",251      "grok-4.20-experimental-beta-non-reasoning-latest",252      "grok-4.20-beta-0309-non-reasoning",253      "grok-4.20-non-reasoning-gv2"254    ],255    "notes": "The only true non-reasoning model in the lineup (fastest TTFT, reasoning_tokens always 0, no reasoning_content). The ONLY model that accepts `stop`. reasoning_effort rejected. frequency/presence_penalty still rejected (400). Good default for cheap/fast chat. Batch API 20% discount."256  },257  {258    "id": "grok-4.20-multi-agent-0309",259    "displayName": "Grok 4.20 Multi-Agent (0309, beta)",260    "family": "grok-4.20",261    "contextTokens": 1000000,262    "maxOutputTokens": null,263    "capabilities": {264      "text": true,265      "vision": true,266      "audioInput": false,267      "audioOutput": false,268      "imageGeneration": false,269      "video": false,270      "reasoning": true,271      "tools": "server-side-only",272      "structuredOutput": null,273      "streaming": true,274      "files": true,275      "webSearch": true276    },277    "parameters": {278      "temperature": null,279      "topP": null,280      "topK": null,281      "maxTokens": false,282      "reasoningEffort": ["low", "medium", "high", "xhigh"],283      "thinkingBudget": false,284      "stop": false,285      "seed": null,286      "frequencyPenalty": false,287      "presencePenalty": false288    },289    "pricing": {290      "inputPerMillion": 1.25,291      "cachedInputPerMillion": 0.2,292      "outputPerMillion": 2.5,293      "longContextThresholdTokens": 200000,294      "longContextInputPerMillion": 2.5,295      "longContextCachedInputPerMillion": 0.4,296      "longContextOutputPerMillion": 5.0,297      "imageInputPerMillionTokens": 1.25298    },299    "status": "beta",300    "aliases": [301      "grok-4.20-multi-agent",302      "grok-4.20-multi-agent-latest",303      "grok-4.20-multi-agent-beta-latest",304      "grok-4.20-multi-agent-experimental-beta-0304",305      "grok-4.20-multi-agent-experimental-beta-latest",306      "grok-4.20-multi-agent-beta-0309"307    ],308    "notes": "/v1/responses ONLY: chat completions returns 400 'Multi Agent requests are not allowed on chat completions' (probed). reasoning.effort selects agent count (low/medium = 4 agents, high/xhigh = 16), not depth. No client-side function tools, no max_tokens. All leader + sub-agent tokens billed -> expensive; lower rate limits (T0 9 RPS / 2.5M TPM). Not probed for generation (cost). Deliberately NOT recommended as a default chat model in PolyLLM."309  },310  {311    "id": "grok-build-0.1",312    "displayName": "Grok Build 0.1",313    "family": "grok-build",314    "contextTokens": 256000,315    "maxOutputTokens": 128000,316    "capabilities": {317      "text": true,318      "vision": true,319      "audioInput": false,320      "audioOutput": false,321      "imageGeneration": false,322      "video": false,323      "reasoning": true,324      "tools": true,325      "structuredOutput": true,326      "streaming": true,327      "files": null,328      "webSearch": null329    },330    "parameters": {331      "temperature": true,332      "topP": true,333      "topK": "accepted-on-chat-completions-undocumented",334      "maxTokens": true,335      "reasoningEffort": false,336      "thinkingBudget": false,337      "stop": false,338      "seed": true,339      "frequencyPenalty": false,340      "presencePenalty": false341    },342    "pricing": {343      "inputPerMillion": 1.0,344      "cachedInputPerMillion": 0.2,345      "outputPerMillion": 2.0,346      "longContextThresholdTokens": 200000,347      "longContextInputPerMillion": 2.0,348      "longContextCachedInputPerMillion": 0.4,349      "longContextOutputPerMillion": 4.0,350      "imageInputPerMillionTokens": 1.0351    },352    "status": "active",353    "aliases": ["grok-code-fast-1", "grok-code-fast", "grok-code-fast-1-0825"],354    "notes": "Cheapest text model; successor of grok-code-fast-1 (those slugs alias here). Always-reasoning with fixed depth: every reasoning_effort value rejected (400 'does not support parameter reasoningEffort'). Tends to spend many reasoning tokens even on trivial prompts (300-650). reasoning_content returned. Coding-oriented; image input accepted (image_tokens billed)."355  }356]357