AI Agent Store/Model comparisonsBuild an AI worker ↗RELEASE WATCH · CHECKED 2026-10-15
Find new models, check which versions have independent results, and estimate how switching could change your bill.
Compare upgrade costs ↓Released 2026-10-14
Matched LiveBench pending
Provider specifications · prices unverified
Choose labels or yes/no answers for records in English, Hindi, Tamil and Kannada. This downloadable model returns probabilities and can serve a ticket-routing workflow on your own hardware.
Provider documentation discovered 2026-10-15
Matched LiveBench pending
Provider specifications · prices unverified
Run the v1.1 decision model on your own GPU to assign labels and probabilities to text or images. This download has different limits from the hosted Gateway offer.
Provider documentation discovered 2026-10-15
Matched LiveBench pending
AI Gateway offer · USD
Sort support tickets, review images or score records against choices you define. This is the new v1.1 Gateway offer; its quality has not been independently compared here.
Provider documentation discovered 2026-10-14
Matched LiveBench pending
AI Gateway offer · USD
Assign text to your own labels, such as sales, support or billing, and receive a probability for each label.
Released 2026-08-11
Matched LiveBench pending
AI Gateway free offer · USD
A newly listed free Gateway offer for NVIDIA’s August model. It supports text assistants, tools and long documents.
Provider documentation discovered 2026-10-14
Matched LiveBench pending
Provider specifications · prices unverified
Google offers a specialized Argon version to approved cyber defenders for finding vulnerabilities and preparing fixes.
Provider documentation discovered 2026-10-13
Matched LiveBench pending
Provider specifications · prices unverified
Run typed decisions on your own hardware. This 35B model uses about 3B parameters per token; it returns option probabilities rather than prose.
Released 2026-10-06
Matched LiveBench pending
Native Drex API · USD
Choose from allowed answers without generating a paragraph: route a ticket, classify a document or return a probability for each option.
Released 2026-10-12
Matched LiveBench pending
Provider specifications · prices unverified
Meta’s higher-capability Wonder variant for work in Muse Code. It requires a Power Usage or Maximum Usage subscription.
Released 2026-10-08
Matched LiveBench pending
Provider specifications · prices unverified
A coding model you can run on your own hardware. Use it to investigate a failing test, draft a fix and check the result while keeping repository access under your control.
Provider documentation discovered 2026-10-11
Matched LiveBench pending
Workers AI documented rates · USD
Choose from your own answer options using text, images, audio or video. Clef Omni returns a confidence score for each answer instead of writing a paragraph.
Provider documentation discovered 2026-10-11
Matched LiveBench pending
Provider specifications · prices unverified
Run this decision model on your own hardware. Give it background information and questions with fixed answer options; it returns confidence scores instead of prose. It is based on Efficient-DLM-8B.
Provider documentation discovered 2026-10-11
Matched LiveBench pending
Provider specifications · prices unverified
Run the downloadable 9B Drex model on your own hardware. It uses the same fixed-answer format as the online service, but has separate hardware, license and setup requirements.
Released 2026-09-28
Matched LiveBench pending
Native Drex API · USD
Send a document to Nace’s online Drex service and ask it to choose an answer, route a ticket or score a condition. It returns the requested decision instead of a paragraph.
Announced 2026-09-28
Matched LiveBench pending
Provider specifications · prices unverified
A coding router that chooses Claude Opus or a Fireworks open model for each user turn. Consider it when routine coding work could run on a cheaper model, then compare completed tasks and total session bills.
Announced 2026-10-06
Matched LiveBench pending
Provider specifications · prices unverified
A text model adapted for Emirati Arabic, including local vocabulary, idioms and cultural context. It can be useful when a general Arabic assistant answers in the wrong dialect.
Announced 2026-10-08
Matched LiveBench pending
Provider specifications · prices unverified
Read page images into text, tables and labeled layout regions. This is the larger checkpoint; test it against the smaller versions on your documents.
Announced 2026-10-08
Matched LiveBench pending
Provider specifications · prices unverified
Read page images into text, tables and labeled layout regions. This variant keeps the earlier LightOnOCR-2 architecture.
Announced 2026-10-08
Matched LiveBench pending
Provider specifications · prices unverified
Read page images into text, tables and labeled layout regions. This is the smallest checkpoint in the family.
Provider documentation discovered 2026-10-10
Matched LiveBench pending
Provider specifications · prices unverified
This second Darwin release adds training for choosing actions and workflow steps. Test it separately from version 1, including whether its confidence scores match how often it is right.
Provider documentation discovered 2026-10-10
Matched LiveBench pending
Provider specifications · prices unverified
A downloadable text model that scores your fixed answer choices in one processing pass per question.
Released 2026-10-09
Matched LiveBench pending
Microsoft base rate / matching Gateway quote · USD
Classify a ticket, choose a workflow step or grade an answer against your rubric. Microsoft’s new text-only decision model returns probabilities over the options you supply, rather than writing a response.
Announced 2026-10-07
Matched LiveBench pending
Provider specifications · prices unverified
A research model that reads and writes bytes instead of ordinary subword tokens. Ai2’s October 7 announcement highlights this new checkpoint; repository creation and modification dates are in August, so announcement and artifact dates stay separate.
Announced 2026-10-07
Matched LiveBench pending
Provider specifications · prices unverified
A research model that reads and writes bytes instead of ordinary subword tokens. Ai2’s October 7 announcement highlights this new checkpoint; repository creation and modification dates are in August, so announcement and artifact dates stay separate.
Released 2026-10-07
Matched LiveBench pending
Provider specifications · prices unverified
An experimental local decision checkpoint for text with images or audio. It can route spoken requests without generating a transcript or free-form answer.
Released 2026-10-07
Matched LiveBench pending
Provider specifications · prices unverified
A local decision checkpoint for text and images. It returns fixed answers in one pass rather than writing a response.
Released 2026-10-05
Matched LiveBench pending
Direct decision API base rate · USD
Use this hosted model when your task has a fixed set of answers, such as routing a ticket or checking a record. It accepts images directly; it is distinct from the two downloadable d1 checkpoints.
Released 2026-10-01
Matched LiveBench pending
AI Gateway offer · USD
A model trained for React Native and React engineering, with Next.js use described by the provider. The native pinned version is October 1; Gateway availability is a separate October 8 offer. Comparable independent coding results are pending.
Released 2026-10-06
Matched LiveBench pending
Direct Decisions API · USD
Use GPT-6 Luna to answer yes/no questions, choose from named options or score an answer against your rules. Each answer has a confidence score. Chat benchmark results do not tell us how accurate these decisions will be.
Released 2026-10-02
Matched LiveBench pending
Provider specifications · prices unverified
A model for turning a research question and supplied literature excerpts into a cited report. Its published development studies do not compare today’s frontier models.
Released 2026-10-03
Matched LiveBench pending
Provider specifications · prices unverified
A German and English reasoning model with tool support and downloadable model files. Comparable independent general-task results and hosted token prices are still missing.
Announced 2026-10-05
Matched LiveBench pending
Provider specifications · prices unverified
A research preview for understanding and generating text, images and video. We have not independently verified general-task quality, response time or billing here.
Released 2026-10-05
Matched LiveBench pending
Provider specifications · prices unverified
A text model for coding and tool-based work, currently behind beta access. Its API limits and five effort settings are documented; comparable independent results and public token prices are not verified.
Provider documentation discovered 2026-10-05
Matched LiveBench pending
Direct restricted API · USD
A specialized life-sciences reasoning model for approved research organizations. OpenAI lists October 5 as its billing start; this is a pricing milestone, not a new model launch. Comparable independent benchmarks are pending.
Announced 2026-09-16
Matched LiveBench pending
AI Gateway decision offer · USD
A decision service from TypeSafe, available through AI Gateway. Gateway announced it on September 16, while its catalog says September 15; neither confirms when the underlying model first launched. Comparable independent accuracy and confidence-score tests are still missing.
Provider documentation discovered 2026-10-03
Matched LiveBench pending
AI Gateway routed offer · USD
AI Gateway offers this version of V4.1 Flash for faster serving. We do not yet have comparable speed measurements, and results for the standard service are not assigned to this version.
Provider documentation discovered 2026-10-03
Matched LiveBench pending
Boundless Gateway promotion · USD
A model for choosing from fixed answers through AI Gateway. Access was announced on October 1. We have not confirmed the underlying model’s launch date or found comparable independent quality tests.
Released 2026-10-01
Matched LiveBench pending
Workers AI documented rates · USD
Use text and images to choose from answers you define. The model files are available under Apache 2.0, and Cloudflare offers an online decision service. We have not independently reproduced the provider’s accuracy or speed comparisons.
Released 2026-10-01
Matched LiveBench pending
Workers AI documented rates · USD
Use text and images to choose from answers you define. The model files are available under Apache 2.0, and Cloudflare offers an online decision service. We have not independently reproduced the provider’s accuracy or speed comparisons.
Provider documentation discovered 2026-10-02
Matched LiveBench pending
Direct Decisions API · USD
A decision model with downloadable Apache 2.0 model files and a dedicated online API. It accepts text and images and gives confidence scores for fixed answers. The provider’s tests are not comparable independent results, and its launch date is not confirmed here.
Provider documentation discovered 2026-10-02
Matched LiveBench pending
AI Gateway paid offer · USD
A text model for coding, analysis and tools. The Gateway free promotion has ended; its standard offer now has published token prices.
Provider documentation discovered 2026-10-01
Matched LiveBench pending
Provider specifications · prices unverified
Provider-documented preview with always-on reasoning and five effort controls; omitted API effort defaults to max. Currently available through M Plan and MiniMax Code. Launch date, public per-token pricing, measured speed and exact matched independent benchmarks remain unverified.
Released 2026-09-17
Matched LiveBench pending
OpenRouter offer · USD
Pareto combines answers from several models for research, coding and multi-step work. This named September version is newly covered here; it is not an October base-model launch.
Catalog added 2026-10-13 UTC
Matched LiveBench pending
OpenRouter offer · USD
Turn a long document into a shorter summary, with a focus or format you specify. Check that important details survive before using it in another workflow.
Catalog added 2026-10-13 UTC
Matched LiveBench pending
OpenRouter offer · USD
Pull fields such as invoice numbers, dates and customer names from documents. Describe the fields you want before sending the text.
Released 2026-10-13
Matched LiveBench pending
OpenRouter offer · USD
A text model for assistants that reason and use tools. You can use a hosted service or download the files to run it on your own hardware.
Announced 2026-09-30
Matched LiveBench pending
OpenRouter offer · USD
Use Argon for demanding coding, research and tasks with several steps. Public API access began October 13. Choose low, medium or high thinking; medium is the default.
Released 2026-10-12
Matched LiveBench
OpenRouter offer · USD
For complex coding and tasks that need several steps. One independent LiveBench result tests max thinking; it does not tell us whether lower settings are faster or better value.
Catalog added 2026-09-04 UTC
Matched LiveBench pending
OpenRouter offer · USD
This health-focused text model’s OpenRouter offer is no longer listed. Current access is unverified, and its earlier domain claims do not establish clinical reliability.
Catalog added 2026-10-08 UTC
Matched LiveBench pending
OpenRouter offer · USD
A new routed offer for StepFun’s coding and knowledge-work model. Native documentation supports text, image and video input, tools, 1M context and 64K output. The Gateway output allowance differs; check your exact endpoint. No matched independent score is assigned.
Released 2026-10-07
Matched LiveBench
OpenRouter offer · USD
Five native effort levels with medium default; thinking is on by default. The 1M context and 128K output limits are documented. Prompts above 100K use five times the base token rates. Its newer tokenizer can count the same text differently from Haiku 4.5; compare actual billed usage before switching.
Compare tested effort levels ↗Released 2026-10-06
Matched LiveBench
OpenRouter offer · USD
Preview API access is available; open weights are planned for later in October. The October 9 routed offer now declares 1,048,576 context tokens, up from 524,288, and 262,144 maximum output. The two-week 50% launch discount announced October 6 is unchanged; exact cutoff time is unverified.
Catalog added 2026-09-21 UTC
Matched LiveBench pending
OpenRouter offer · USD
Switchyard chooses from a configured group of models. OpenRouter says each request uses the selected model’s price and input capacity. There is no single confirmed fixed rate or guaranteed one-million-token capacity for every request; catalog placeholders do not mean it is free.
Released 2026-10-01
Matched LiveBench pending
OpenRouter offer · USD
The provider describes a system that combines models, rather than a separate model you can download. Its preview can change. The provider’s early tests use different questions and tools, so we do not assign a comparable benchmark score here.
Catalog added 2026-10-01 UTC
Matched LiveBench pending
OpenRouter offer · USD
This reasoning and tool-use model’s free OpenRouter offer is no longer listed. Confirm another available offer before connecting an application.
Released 2026-09-29
Matched LiveBench
OpenRouter offer · USD
Five supported effort levels: low, medium (the API default), high, extra high and max. None and minimal are unsupported. Tool calling uses Responses; Chat Completions supports requests without tools.
Compare tested effort levels ↗Released 2026-09-28
Matched LiveBench
OpenRouter offer · USD
Adaptive thinking, high default effort and 128K maximum output. Migration changes affect forced tool use, thinking replay and older computer-use tools.
Compare tested effort levels ↗Catalog added 2026-09-29 UTC
Matched LiveBench pending
OpenRouter offer · USD
OpenRouter describes this as GPT-6.1 Sol served in pro mode, with greater token use. It is a serving variant; no independent result is assigned from the base model.
Catalog added 2026-09-23 UTC
Matched LiveBench pending
OpenRouter offer · USD
A separately listed service for GLM 5.3. We have not verified its advertised speed with comparable measurements or assigned it the standard model’s benchmark results.
Catalog added 2026-09-23 UTC
Matched LiveBench pending
OpenRouter offer · USD
A separately listed service for Qwen3.8 Max. Comparable benchmark results and measured speed for this specific version are not confirmed.
Released 2026-09-23
Matched LiveBench pending
OpenRouter offer · USD
Research preview announced September 23 with temporary two-week serverless access. Vendor claims about shorter reasoning are not independently measured task costs; verify availability.
Released 2026-09-23
Matched LiveBench pending
OpenRouter offer · USD
A text-generation model listed in the provider catalog. We have not verified its general-task or storytelling performance in comparable independent tests.
Released 2026-09-23
Matched LiveBench pending
OpenRouter offer · USD
A lower-priced text-generation service. Test answer quality and retries before assuming it will cost less to complete your tasks.
Released 2026-09-25
Matched LiveBench pending
OpenRouter offer · USD
Provider verifies text, image, audio and video input and 32K context; this catalog offer advertises 36,864 tokens. Confirm the selected endpoint limit. Specialized performance is not scored using another model’s text benchmarks.
New: compare decision models for routing and classification ↗
Compare Standard, Fast and Ultrafast costs for GPT-6.1 Sol and Astra ↗
Check model routes that disappeared from a catalog ↗
Data correction · nvidia/switchyard. OpenRouter corrected the previous free-service FAQ and clarified that context is the serving model’s limit. The routing page preserves unknown fixed prices and exact system identity.
Data correction · mistral-large-4-high. LiveBench revised six task values in the existing Large 4 high row. Today’s category scores use the corrected row; the October 7 download preserves its original values. This is a benchmark correction, not a new model release or effort sweep.
Data correction · @cf/cloudflare/clef-flash and @cf/cloudflare/clef. Current exact Workers AI pages correct stale profile facts: Flash is24,576tokens/$0.038input, and both Clef endpoints explicitly charge no output tokens. This is a source-backed correction to our data; a provider price-change date is unverified. Immutable prior downloads retain the old observation.
SAME SERVICE · PRICES CHECKED TWICE
A service such as OpenRouter can change its price without the model maker changing its own price. These quotes show what we found on each date, per million tokens. Prompt length, time of use, the provider handling the request and other fees can affect your bill.
2026-10-14 → 2026-10-15
Prices for this routed offer changed. Check the service you plan to use; direct-provider prices may differ.
Check the current offer ↗2026-10-14 → 2026-10-15
Prices for this routed offer changed. Check the service you plan to use; direct-provider prices may differ.
Check the current offer ↗2026-10-14 → 2026-10-15
Prices for this routed offer changed. Check the service you plan to use; direct-provider prices may differ.
Check the current offer ↗2026-10-14 → 2026-10-15
Prices for this routed offer changed. Check the service you plan to use; direct-provider prices may differ.
Check the current offer ↗2026-10-14 → 2026-10-15
Prices for this routed offer changed. Check the service you plan to use; direct-provider prices may differ.
Check the current offer ↗2026-10-14 → 2026-10-15
Prices for this routed offer changed. Check the service you plan to use; direct-provider prices may differ.
Check the current offer ↗2026-10-14 → 2026-10-15
Prices for this routed offer changed. Check the service you plan to use; direct-provider prices may differ.
Check the current offer ↗ACCESS AND BILLING
Deprecated; retirement scheduled
claude-sonnet-4-5-20250929
Provider-recommended replacement: claude-sonnet-5-5.
Claude API retirement notice. Partner-operated Bedrock and Google Cloud schedules can differ. Historical benchmarks remain historical; scores are not transferred to the recommended replacement.
Read the provider notice ↗Catalog expiration scheduled
z-ai/glm-4.5
Replacement recommendation: Not verified.
The exact OpenRouter entry now declares this expiration date. This is endpoint-specific catalog evidence; it does not establish a direct-provider shutdown or retirement of every hosted copy. No exact cutoff time or replacement is verified.
Read the endpoint catalog ↗Catalog expiration scheduled
z-ai/glm-4.7
Replacement recommendation: Not verified.
The exact OpenRouter entry now declares this expiration date. This is endpoint-specific catalog evidence; it does not establish a direct-provider shutdown or retirement of every hosted copy. No exact cutoff time or replacement is verified.
Read the endpoint catalog ↗Catalog expiration scheduled
dots-studio/dots-3-note-preview:free
Replacement recommendation: Not verified.
The exact OpenRouter entry now declares this expiration date. This is endpoint-specific catalog evidence; it does not establish a direct-provider shutdown or retirement of every hosted copy. No exact cutoff time or replacement is verified.
Read the endpoint catalog ↗Deprecated; retirement scheduled
gpt-5.3-codex
Provider-recommended replacement: gpt-6-sol.
Official OpenAI API notice gives six months of notice. Historical results remain attached to the tested model; no successor scores are transferred.
Read the provider notice ↗Deprecated; retirement scheduled
gpt-5.4-nano
Provider-recommended replacement: gpt-6-luna.
Official OpenAI API notice gives six months of notice. Historical results remain attached to the tested model; no successor scores are transferred.
Read the provider notice ↗Deprecated; retirement scheduled
gpt-5.1
Provider-recommended replacement: gpt-6-sol.
Official OpenAI API notice gives six months of notice. Historical results remain attached to the tested model; no successor scores are transferred.
Read the provider notice ↗Hosted endpoint deprecated; retirement scheduled
zai-glm-5-2
Provider-recommended replacement: zai-glm-5-3.
This notice applies to the Mistral-hosted endpoint. It does not retire every Z.ai GLM5.2 deployment. The provider states the replacement uses the same price; no benchmark score is transferred.
Read the provider notice ↗Managed agent · earliest shutdown date reached
antigravity-preview-05-2026
Provider-recommended replacement: antigravity-preview-09-2026.
Google lists October 5 as the earliest shutdown date for this managed agent, rather than a base model. We have not tested whether the endpoint is already unavailable. Local-tool integrations need the new parameter and file-edit contracts described in the September 17 release notes; changing only the agent ID may be insufficient.
Read the provider notice ↗Mutable alias · snapshot updated
chat-latest
Provider-recommended replacement: Not verified.
OpenAI updated this chat snapshot on October 7. It remains a moving alias, not a newly named base model or a pinned benchmark version. The provider recommends the GPT-6 family for production APIs; no predecessor score is transferred.
Read the provider notice ↗Routed alias · now points to Haiku 5.5
~anthropic/claude-haiku-latest
Replacement recommendation: anthropic/claude-haiku-5.5.
The exact catalog alias changed from Haiku 4.5 to Haiku 5.5 between the October 7 and October 8 observations. Thinking defaults, tokenization and long-context prices change with it. Pin and test a specific model before depending on this alias; native Anthropic IDs are separate contracts.
Read the endpoint catalog ↗Deprecated; shutdown scheduled
deep-research-pro-preview-12-2025
Provider-recommended replacement: deep-research-preview-04-2026 or deep-research-max-preview-04-2026.
This applies to the managed Deep Research agent, not the underlying Gemini model family. Update the agent parameter in interactions.create and test speed versus comprehensiveness on your research tasks.
Read the provider notice ↗Gateway offer returned October 13
deepseek/deepseek-v4.1-flash-fast
Provider-recommended replacement: Not verified.
This route returned October 13 after its October 9 absence. Current output and prices differ from the previous offer; see the current profile before using it.
Read the provider notice ↗Native alias and retired-ID fallback
drex-latest / drex-v1.1
Provider-recommended replacement: drex-v2.0 / drex-v1.5.
Since October 12, omitted model and drex-latest select v2.0. Retired drex-v1.1 still selects v1.5. Current v2.0 refuses requests above its limits; the previously documented long-request fallback has been removed.
Read the provider notice ↗Native cache-price change in effect
step-5-preview
Since October 14 at 00:30 Beijing time, the standard native cache-read price increased from ¥0.35 to ¥0.70 per million tokens. Users with a pay-as-you-go call before the cutoff keep ¥0.35. Step Plan is excluded. Uncached input ¥7 and output ¥20 remain unchanged. These native yuan rates do not replace routed dollar quotes.
Read the provider notice ↗Deprecated; retirement scheduled
gpt-4o-2024-08-06
Provider-recommended replacement: gpt-6-luna.
OpenAI plans to remove this API snapshot on January 14, 2027. Fine-tuned versions follow the same schedule. Test the replacement on your own work; old benchmark scores do not transfer.
Read the provider notice ↗Deprecated; retirement scheduled
gpt-4o-2024-11-20
Provider-recommended replacement: gpt-6-luna.
OpenAI plans to remove this API snapshot on January 14, 2027. Fine-tuned versions follow the same schedule. Test the replacement on your own work; old benchmark scores do not transfer.
Read the provider notice ↗Deprecated; retirement scheduled
gpt-4o-mini-2024-07-18
Provider-recommended replacement: gpt-6-luna.
OpenAI plans to remove this API snapshot on January 14, 2027. Fine-tuned versions follow the same schedule. Test the replacement on your own work; old benchmark scores do not transfer.
Read the provider notice ↗Deprecated; retirement scheduled
gpt-4.1-2025-04-14
Provider-recommended replacement: gpt-6-luna.
OpenAI plans to remove this API snapshot on January 14, 2027. Fine-tuned versions follow the same schedule. Test the replacement on your own work; old benchmark scores do not transfer.
Read the provider notice ↗Deprecated; retirement scheduled
gpt-4.1-mini-2025-04-14
Provider-recommended replacement: gpt-6-luna.
OpenAI plans to remove this API snapshot on January 14, 2027. Fine-tuned versions follow the same schedule. Test the replacement on your own work; old benchmark scores do not transfer.
Read the provider notice ↗Expiration date reached; still catalog-listed
unbiased/pareto-26.10-preview
Replacement recommendation: Not verified.
The October 15 catalog still lists this preview with an October 15 expiration. Confirm access before relying on it; a scheduled date alone does not prove that requests already fail.
Read the endpoint catalog ↗Moving alias · points to Pareto 26.9
~unbiased/pareto-latest
Replacement recommendation: unbiased/pareto-26.9.
This alias currently selects Pareto 26.9. It can move to another version without changing your request ID; use a named version when repeatability matters.
Read the endpoint catalog ↗Introductory prices end November 30
gemini-4-argon
Standard input/cache/output prices per million tokens change from $2/$0.10/$10 to $4/$0.20/$20 on December 1. Batch, Flex and Priority have separate rates; thinking is billed as output.
Read the provider notice ↗Documented model ID changed
Beam-1.0-501B-A23B
Provider-recommended replacement: Not verified.
Current examples use Beam-1.0-501B-A23B instead of Beam-501B-A23B. Check the models available to your account. A new spelling is not evidence of a new model checkpoint or a retirement date.
Read the provider notice ↗SAME TOKEN ASSUMPTIONS · DIFFERENT BILL
Pick two models and enter your monthly usage. The estimate uses OpenRouter offers checked 2026-10-15.
Tokens are the pieces of text a model processes. Include charged thinking tokens in output. Caching reuses previously stored input, so those tokens are already part of your total input. This comparison uses the same counts for both models; actual usage, answer quality and caching can differ.
Catalog standard token rates.
Check this offer ↗Catalog standard token rates.
Check this offer ↗Same estimate with these token assumptions.
The same text can use more tokens with a new tokenizer. Haiku 5.5 counts about 30% more than Haiku 4.5 on average, so repeat this estimate with your measured usage. Excludes cache writes, retries, tool fees, image/audio-specific billing, hosting, taxes and service tiers. The caching estimate assumes your reusable input is already stored and qualifies for a discount. Rules differ by service. Only confirmed catalog prices for long prompts are included. A large input limit does not guarantee the model will use every detail correctly.
Claude Haiku 5.5, GPT-6.1 Sol and Claude Sonnet 5.5 have extra high and max results on the same LiveBench question set. Lower settings remain unmeasured in this dataset. Their dollar costs and durations are not published here.
A daily review is scheduled for 05:30 Europe/Vilnius. This page shows the last completed review, not a live feed. A source checked today can still contain older results. Missing independent evidence stays pending.
Reviewed text models, serving offers, composite systems and specialized decision and research models discovered since the September 23 snapshot. These are distinct kinds of systems, not a count of independent base models. Discovery and catalog addition dates are not verified launch dates. Upgrade cost scenarios use selected OpenRouter offers; other documented prices are shown separately.
June25 data and categories unchanged; v2 current bundle ba5ba12f inspected separately, repository404 and reuse/setup unresolved. No v2 numeric import.
Check source ↗HTTP403; retained October13 extraction and October8 publisher cutoff.
Check source ↗Reuse restrictions retained; historical observations unchanged.
Check source ↗Primary sources checked October 15. No additional covered release verified; original measurement dates retained.
Check source ↗462 routes; no new IDs; unversioned unbiased/pareto absent.
Check source ↗Primary sources checked October 15. No additional covered release verified; original measurement dates retained.
Check source ↗Primary sources checked October 15. No additional covered release verified; original measurement dates retained.
Check source ↗Contributor private-preview pricing added with product-improvement data use; standard quotes unchanged.
Check source ↗Primary sources checked October 15. No additional covered release verified; original measurement dates retained.
Check source ↗October14GrokVoice3.0 speech-to-speech release noted outside text comparison; no new general text model verified.
Check source ↗Primary sources checked October 15. No additional covered release verified; original measurement dates retained.
Check source ↗Primary sources checked October 15. No additional covered release verified; original measurement dates retained.
Check source ↗Primary sources checked October 15. No additional covered release verified; original measurement dates retained.
Check source ↗Primary sources checked October 15. No additional covered release verified; original measurement dates retained.
Check source ↗Primary sources checked October 15. No additional covered release verified; original measurement dates retained.
Check source ↗Primary sources checked October 15. No additional covered release verified; original measurement dates retained.
Check source ↗Primary sources checked October 15. No additional covered release verified; original measurement dates retained.
Check source ↗Primary sources checked October 15. No additional covered release verified; original measurement dates retained.
Check source ↗Primary sources checked October 15. No additional covered release verified; original measurement dates retained.
Check source ↗Bare firerouter chooses Opus or Sol from credentials/client; explicit opus profile and fallback/billing clarified.
Check source ↗Primary sources checked October 15. No additional covered release verified; original measurement dates retained.
Check source ↗Primary sources checked October 15. No additional covered release verified; original measurement dates retained.
Check source ↗Primary sources checked October 15. No additional covered release verified; original measurement dates retained.
Check source ↗Unversioned OpenRouter offer absent;26.10preview stilllisted despiteOctober15expiry. No universal retirement inferred.
Check source ↗New v1.1 Gateway offer and Apache2 local checkpoint verified separately; 262144 hosted versus8192 supplied local implementation. Vendor evaluations not imported as independent quality.
Check source ↗Primary sources checked October 15. No additional covered release verified; original measurement dates retained.
Check source ↗No new licensed matched multi-effort study verified.
Check source ↗465 routes; new pplx-decider-v1.1-27b verified with hosted quote separate from download.
Check source ↗Primary sources checked October 15. No additional covered release verified; original measurement dates retained.
Check source ↗Reuse restrictions retained; no numeric import.
Check source ↗Primary sources checked October 15. No additional covered release verified; original measurement dates retained.
Check source ↗Primary sources checked October 15. No additional covered release verified; original measurement dates retained.
Check source ↗Licensed14completeUTCdays October1–14 refreshed, separate text/all populations.
Check source ↗Documented ID nowBeam-1.0-501B-A23B; logprobs output cap8192 including reasoning and plan limit. Not counted as new base model.
Check source ↗Primary sources checked October 15. No additional covered release verified; original measurement dates retained.
Check source ↗Primary sources checked October 15. No additional covered release verified; original measurement dates retained.
Check source ↗Primary sources checked October 15. No additional covered release verified; original measurement dates retained.
Check source ↗Primary sources checked October 15. No additional covered release verified; original measurement dates retained.
Check source ↗Primary sources checked October 15. No additional covered release verified; original measurement dates retained.
Check source ↗Primary sources checked October 15. No additional covered release verified; original measurement dates retained.
Check source ↗Primary sources checked October 15. No additional covered release verified; original measurement dates retained.
Check source ↗Primary sources checked October 15. No additional covered release verified; original measurement dates retained.
Check source ↗Primary sources checked October 15. No additional covered release verified; original measurement dates retained.
Check source ↗Primary sources checked October 15. No additional covered release verified; original measurement dates retained.
Check source ↗Primary sources checked October 15. No additional covered release verified; original measurement dates retained.
Check source ↗Primary sources checked October 15. No additional covered release verified; original measurement dates retained.
Check source ↗Primary sources checked October 15. No additional covered release verified; original measurement dates retained.
Check source ↗Primary sources checked October 15. No additional covered release verified; original measurement dates retained.
Check source ↗Primary sources checked October 15. No additional covered release verified; original measurement dates retained.
Check source ↗Primary sources checked October 15. No additional covered release verified; original measurement dates retained.
Check source ↗v2 native now131072state/262144row, refuses over-limit instead of v1.5 fallback; body cap4MiB. Prices unchanged.
Check source ↗Primary sources checked October 15. No additional covered release verified; original measurement dates retained.
Check source ↗Primary sources checked October 15. No additional covered release verified; original measurement dates retained.
Check source ↗Native Extract now documents at most30distinctvalues perentitytype; pricing unchanged.
Check source ↗Lumma-Fev-27B October14 release added as local decision checkpoint; provider claims kept separate.
Check source ↗October14 UltraBERT150M/380M andEmbed150M verified as encoders/embeddings, outside general-chat and fixed-choice decision comparison; no scores mixed.
Check source ↗Download release observations ↓ · Current model benchmark snapshot ↓ · Current effort snapshot ↓