Skip to content
ReviewRamble

The weekly brief

AI, reviewed, in your inbox.

  • New reviews of tools, models, MCP servers and repos, with the verdict up front
  • Re-checked prices and ratings when something changes
  • New analysis pieces as they are published

Models, agents, MCP servers and the tools actually worth your time.

One email a week. Unsubscribe with the link in any brief. How we use your email

The best large language models (LLMs) in 2026

The best large language models (LLMs) right now are Claude Opus 5.5, Claude Sonnet 5.5 and GPT-6.1 Sol, out of 44 we have reviewed. Each is ranked by our rating, with its price, our verdict and who it suits.

By ReviewRamble editors · Updated Nov 18, 2026 · How we review

How we ranked: by our rating out of 5, from reviews built on each product's own pricing, documentation and policies plus independent benchmarks where they exist, each checked on the date shown in the review. We take no affiliate commission.

  1. 1

    Anthropic's new default earns the title: first on Artificial Analysis's independent index, and a fifth cheaper per token than the Opus it replaces.

    Best for: Teams running long coding agents, code audits or document-heavy research where finishing the task matters more than the cost of each call.

    Price: API: $4 per 1M input tokens, $20 per 1M output; cache reads $0.20, 5-minute cache writes $5; Batch API 50% off; fast mode $8/$40; US-only inference 1.1x.

    Read our Claude Opus 5.5 review
    4.7/5
  2. 2

    Within two points of Opus 5.5 on Anthropic's knowledge-work test at half the price, which turns Opus into a deliberate upgrade rather than the default.

    Best for: Teams running coding agents, support automation or document work at volume who want near-Opus quality at half the token price.

    Price: API: $2 per 1M input tokens, $10 per 1M output tokens; cache reads $0.10; Batch API 50% off.

    Read our Claude Sonnet 5.5 review
    4.6/5
  3. 3

    For most OpenAI customers this is the flagship that matters: a point behind Astra on Artificial Analysis's index at a fifth of the token price.

    Best for: Teams building agents and business workflows on OpenAI who want near-Astra results at mid-tier cost and can live with slower output.

    Price: API: $2 per 1M input tokens, $10 per 1M output; cached input $0.10; Batch/Flex half price; prompts over 272K input tokens pay 2x input and 1.5x output for the whole request.

    Read our GPT-6.1 Sol review
    4.6/5
  4. 4

    Haiku 5.5 outscores Gemini 3.8 Flash on independent tests at a sixth of the cost per task, which resets what anyone should expect from a budget model.

    Best for: High-volume classification, extraction, support and subagent work on short prompts.

    Price: API: $0.10 input / $0.50 output per 1M tokens for prompts up to 100K tokens; $0.50 / $2.50 above that; cache reads $0.01; Batch API 50% off.

    Read our Claude Haiku 5.5 review
    4.5/5
  5. 5

    A 27B model with an Apache licence that, on Alibaba's own numbers, edges past Claude Opus 4.6 Max at coding and fits on a single 24 GB card at 4-bit.

    Best for: Developers who want the strongest permissively licensed model that fits on one high-end GPU, especially for coding and computer-use agents.

    Price: Free weights under Apache 2.0. Qwen Cloud: $0.50 input / $3.00 output per 1M tokens ($0.10 cached). Third-party hosts on OpenRouter from about $0.04 / $1.25.

    Read our Qwen3.8-27B review
    4.5/5
  6. 6

    Hydrogen edges Argon on Artificial Analysis for under 40% of Argon's December price, so it should be the default Gemini model for routine work; Claude Sonnet 5.5 still wins on terminal and computer-use tasks.

    Best for: Gemini API and Antigravity users who want Argon-class results for routine coding and agent steps at well under Carbon's price.

    Price: API: $1.50 input / $7.50 output per 1M tokens, flat to the full 1M window; cached input $0.075; Batch and Flex $0.75 / $3.75; Priority $2.70 / $13.50. Free tier in AI Studio (data used to improve Google products). Antigravity: paid Google AI Pro or Ultra.

    Read our Gemini 4 Hydrogen review
    4.4/5
  7. 7

    GLM-5.3-Flash gets most of GLM-5.3's coding ability for a tenth of the token price, and its MIT licence leaves nothing for a lawyer to read.

    Best for: Teams running high-volume coding or multimodal agents who want an open model with no licence conditions and very low API costs.

    Price: Open weights under MIT. Z.ai API: $0.15 input / $0.50 output per 1M tokens ($0.03 cached); FlashX $0.37 / $1.25.

    Read our GLM-5.3-Flash review
    4.3/5
  8. 8

    Luna scores five points below Claude Haiku 5.5 on Artificial Analysis's index and costs a third as much per task, which is the whole decision in one line.

    Best for: OpenAI-based apps doing classification, extraction or chat at very high volume.

    Price: API: $0.10 per 1M input tokens, $0.50 per 1M output tokens; cached input $0.01; long prompts over 272K cost more.

    Read our GPT-6 Luna review
    4.3/5
  9. 9

    Near-frontier coding-agent scores for $0.15 per million input tokens, provided you can accept China-based data handling or host a 763B-parameter checkpoint yourself.

    Best for: Cost-sensitive teams running coding and automation agents who can use a China-hosted API or a third-party host, or have multi-GPU servers for the open weights.

    Price: Free MIT-licensed weights. DeepSeek API: $0.15 input / $0.60 output per 1M tokens off-peak, $0.30 / $1.20 at weekday peak; cache hits $0.003 to $0.006.

    Read our DeepSeek-V4.1-Flash review
    4.2/5
  10. 10

    Muse Ember is the cheapest way to get Haiku-class agent work done, provided you do not need it to know things.

    Best for: Teams running high-volume agent subtasks, extraction and routing who already pay for a larger planner model.

    Price: Up to 200K input tokens: $0.10 input, $0.004 cached input, $0.50 output per 1M tokens. Over 200K: $0.30 / $0.012 / $1.50, applied to the whole request. Web search $2.50 per 1,000 queries.

    Read our Muse Ember review
    4.2/5
  11. 11

    A free speed-up for anyone serving Qwen3.8-27B: roughly three times faster generation on agent workloads, with output identical to the base model, if you run SGLang.

    Best for: Teams serving Qwen3.8-27B with SGLang who want lower latency per request without changing the model's answers.

    Price: Free open weights under Apache 2.0; you pay only for the GPUs that serve Qwen3.8-27B.

    Read our Qwen3.8-27B DSpine review
    4.2/5
  12. 12

    Strong agent and coding scores for 6 billion active parameters, under a licence that is stricter than the one on Qwen's own flagship.

    Best for: Teams running internal coding or agent workloads that outgrow a 27B model, and developers who want a very cheap hosted multimodal API.

    Price: Open weights under the Qwen Community License 1.0 (separate licence for hosting and AI-assistant businesses). Qwen3.8-Flash API: $0.15 input / $0.47 output per 1M tokens; cache hits $0.016.

    Read our Qwen3.8-Flash-Next review
    4.2/5
  13. 13

    The strongest Chinese model for video and visual understanding on ByteDance's own numbers, but closed, unscored independently and hard to buy outside China.

    Best for: Teams that need strong video and image understanding and can buy through ByteDance's Volcano Engine in China, or use Seed 2.1 Turbo internationally.

    Price: Volcano Engine Ark (China): Pro ¥6 input / ¥30 output per 1M tokens, cached ¥1.20, batch ¥3 / ¥15; Turbo half that. BytePlus ModelArk (international): Turbo $0.50 / $2.50 per 1M tokens; Pro not in the published list.

    Read our Doubao Seed 2.1 Pro review
    4.1/5
  14. 14

    Quick to stream and broad in what it reads, but Haiku 5.5 and GPT-6 Luna now charge a third of its input price and a fifth of its output.

    Best for: Teams on Google Cloud running high-volume extraction or audio and video understanding where time to first token does not matter.

    Price: API: $0.30 input / $2.50 output per 1M tokens; cached input $0.03; Batch and Flex $0.15 / $1.25; free tier (prompts may be used for training).

    Read our Gemini 3.5 Flash-Lite review
    4.1/5
  15. 15

    Broad inputs and a free tier keep Gemini 3.8 Flash useful, but it now spends more per task than rivals that score higher, and its price doubles in January.

    Best for: Teams processing audio, video and long PDFs inside Google's ecosystem who can tolerate slow first responses.

    Price: API: $0.75 input / $3.75 output per 1M tokens until 31 Dec 2026, then $1.50 / $7.50; Batch and Flex half price; free tier with limits.

    Read our Gemini 3.8 Flash review
    4.1/5
  16. 16

    Near-Sol results at half the price make it OpenAI's pick for agents that click and type, though its system card says to keep a human on destructive steps.

    Best for: Teams running high-volume computer-use or office agents who want results close to Sol's at half the price and can add approval steps.

    Price: API: $1 per 1M input tokens, $5 per 1M output; cached input $0.05; Batch and Flex half price; Fast mode 2x; prompts over 272K input tokens pay 2x input and 1.5x output for the whole request; regional processing 10% extra.

    Read our GPT-6.1 Terra review
    4.1/5
  17. 17

    Europe's trillion-parameter model is now one anyone can download and run in-house, at half its preview price, though independent tests still place it mid-pack among big open models.

    Best for: European organisations that want a large multilingual model they can run on their own Blackwell or H200 cluster, or call cheaply through Mistral's API.

    Price: Open weights under a modified MIT licence (a separate licence is needed to sell it as a hosted API). Mistral API: $0.70 per 1M input tokens ($0.07 cached), $2.10 per 1M output; batch 50% off.

    Read our Mistral Large 4 review
    4.1/5
  18. 18

    Glimmer is the open model most carefully fitted to the graphics card people own, and Qwen3.8-27B beat it on the benchmarks within weeks.

    Best for: Developers building local, always-on agents on one consumer GPU who want a permissive licence from a US lab.

    Price: Free weights under Apache 2.0; no Meta API. Hosted on OpenRouter from $0.30 per 1M input and $1.20 per 1M output tokens (DeepInfra).

    Read our Muse Glimmer review
    4.1/5
  19. 19

    Wonder costs what Claude Opus 5.5 costs and wins only five of the fourteen rows Meta chose to publish against it.

    Best for: Teams already on Meta's API that need a stronger model for long, multi-file coding or legal document work than Muse Spark provides.

    Price: API: $4 input, $0.20 cached input, $20 output per 1M tokens; no Contributor mode. Wonder Full only in Muse Code Power ($50/month) and Maximum ($200/month).

    Read our Muse Wonder review
    4.1/5
  20. 20

    A strong pretrained base model that leads its class on Baidu's own table, useful mainly to teams who fine-tune their own models rather than call an API.

    Best for: Teams that pretrain or fine-tune their own models and want a fast, knowledgeable mixture-of-experts base rather than a ready-to-use chat model.

    Price: Free weights under Baidu's custom ERNIE 5.5 Flash Base licence. No hosted price for the base model; you pay the GPU to run or fine-tune it.

    Read our ERNIE 5.5 Flash Base review
    4.0/5
  21. 21

    Near the top of the open-weight table for a fraction of a cent per task, as long as nothing it says goes out unchecked.

    Best for: Teams running high-volume agent and automation work, or long multimodal inputs, who can verify outputs and accept MiniMax's licence terms.

    Price: Open weights under the MiniMax community licence. MiniMax API per 1M tokens, flat at any prompt length: $0.15 input, $0.60 output, $0.015 cached input.

    Read our MiniMax-M3.1-Flash review
    4.0/5
  22. 22

    Cheap, long-context and well-mannered in agent loops, but squeezed from above by Muse Wonder and from the side by an open Spark at the same price.

    Best for: Teams that want a cheap, long-context model for sub-agents and tool loops and are content to send prompts to Meta's API.

    Price: Standard: $1.25 input, $0.15 cached input, $4.25 output per 1M tokens. Contributor mode: $0.10 / $0.002 / $0.20, and Meta may train on your content.

    Read our Muse Spark 1.3 review
    4.0/5
  23. 23

    A serious open-weight agent model with unusually strong safety results, held back by hosted prices that match closed Spark and a licence that asks for your birth date.

    Best for: Organisations that need a capable agent model on their own hardware, with stronger refusal behaviour than other open models, and can live with Meta's attribution clause.

    Price: Free weights under the Muse Model License (attribution required; $20M revenue threshold for model-as-a-service). Hosted on OpenRouter at $1.25 input, $0.15 cached, $4.25 output per 1M tokens.

    Read our Muse Spark Open review
    4.0/5
  24. 24

    Strong for its price on long, mixed-media inputs, held back by a licence that makes lawyers read twice.

    Best for: Cost-sensitive teams processing long documents or video who want an open model and can accept its licence terms.

    Price: Open weights under the MiniMax community licence. API per 1M tokens on prompts up to 512K: $0.30 input, $1.20 output, $0.06 cached; double above 512K; priority tier 1.5x.

    Read our MiniMax-M3 review
    3.9/5
  25. 25

    Apache 2.0 weights, two-H100 hosting and every EU language make Command A+ easy to approve, but Cohere still publishes no benchmark scores to judge it by.

    Best for: Enterprises building private, multilingual RAG and agent systems who want a permissively licensed model they can host on two GPUs.

    Price: Free weights under Apache 2.0. Cohere API rate not published; OpenRouter lists $0.30 input / $1.50 output per 1M tokens. Private deployment by quote.

    Read our Command A+ review
    3.8/5
  26. 26

    Cheap on the rate card and dear in practice: Grok 4.7 thinks for so long that its cost per task lands among the highest measured.

    Best for: Developers who need live X search inside an agent and can tolerate long thinking times, or teams already committed to SpaceXSI's platform.

    Price: API: $2 input / $6 output per 1M tokens under 200K-token prompts, $4 / $12 at or above; cached input $0.50; US endpoint 10% extra.

    Read our Grok 4.7 review
    3.8/5
  27. 27

    Hy4 is a cheap, capable second tier: vision and computer use at preview prices, but it leads none of Tencent's own benchmarks and its open weights are still a promise.

    Best for: Teams that want a low-cost agent model with screen control and image input, and can wait for weights before self-hosting.

    Price: OpenRouter (Tencent) list price $0.834 input / $2.501 output per 1M tokens, cached input $0.042; Tencent quotes about $0.90 / $2.70. TokenHub is 50% off until 31 December 2026. Weights not yet released.

    Read our Hy4 review
    3.8/5
  28. 28

    A cheap, MIT-licensed medical specialist that beats Claude Opus 4.8 on Ant's own HealthBench Professional run, though only Ant has measured it and it still trails GPT-5.6 Sol.

    Best for: Health-tech and research teams that want an open, permissively licensed medical model to self-host behind human review.

    Price: Open weights under MIT. Vercel AI Gateway: $0.07 input / $0.22 output per 1M tokens (Novita's row: $0.08 input, $0.02 cached).

    Read our Ling-3.0-flash-Sante review
    3.8/5
  29. 29

    A trillion-scale open model under a plain MIT licence, cheap to call but already overtaken by its own successor.

    Best for: Developers who want a cheap, MIT-licensed large model for coding agents and can live with a single hosting provider.

    Price: Open weights under the MIT licence. OpenRouter (AtlasCloud): $0.30 input / $0.006 cached / $1.20 output per 1M tokens.

    Read our LongCat-2.0 review
    3.8/5
  30. 30

    A capable four-GPU coding model whose licence excludes large companies and whose API price now loses to Mistral's own Large 4.

    Best for: Companies under $20 million in monthly revenue that want to self-host a strong coding and agent model on one four-GPU server.

    Price: Open weights under a modified MIT licence (no rights for companies above $20M monthly revenue). API: $1.50 per 1M input tokens, $7.50 per 1M output; batch 50% off.

    Read our Mistral Medium 3.5 review
    3.8/5
  31. 31

    Arete-27B shows that a Chinese model's political framing comes off with one modest fine-tune, but it is a Qwen3.8-27B with a different worldview, not a better Qwen.

    Best for: Teams running Qwen3.8-27B locally who need it to answer politically sensitive questions without Chinese state framing.

    Price: Free Apache 2.0 weights. No first-party API; you pay for your own hardware.

    Read our Arete-27B review
    3.7/5
  32. 32

    Cheap to train and good at search-backed answers and maths, but closed, text-only on the API and still hard to buy outside China.

    Best for: Teams building for Chinese-language users on Baidu's cloud who want strong search-grounded answers at a low token price.

    Price: Qianfan (mainland China): ¥0.004 input / ¥0.018 output per 1,000 tokens up to 32K, ¥0.006 / ¥0.022 from 32K to 128K. No official dollar price; the international catalogue lists ERNIE 5.0 at $1.40 / $5.60 per 1M tokens. Free in the ERNIE assistant.

    Read our ERNIE 5.1 review
    3.7/5
  33. 33

    A narrow model with a sound idea, writing agent memory as well as Claude Opus 5 on Tencent's tests at a fraction of the cost, though only Tencent has measured it.

    Best for: Teams running Mem0, MemOS or similar memory layers at volume who want to replace a paid frontier model on the write path with self-hosted weights.

    Price: Open weights under Apache 2.0; no hosted API.

    Read our Hy-Memory-30B-A3B review
    3.7/5
  34. 34

    On its makers' own tests Maxwell is the strongest 3B-active model for agentic search, but every number is self-reported and the company behind it is not named.

    Best for: Teams building search and research agents who can host a roughly 74 GB model and want to test a 3B-active alternative to Qwen3.6-35B-A3B.

    Price: Free open weights under Apache 2.0. A web playground is offered; no paid API or pricing is published.

    Read our Maxwell-V1.0-Flash review
    3.7/5
  35. 35

    Cheap, fast and cleanly licensed, but independent tests put it well behind the open models it shares a server with.

    Best for: Teams wanting a low-cost, Apache-licensed model for chat, documents and images, on Mistral's API or their own GPUs.

    Price: Free weights under Apache 2.0. API: $0.15 input / $0.60 output per 1M tokens; batch 50% off; cached input up to 90% off.

    Read our Mistral Small 4 review
    3.7/5
  36. 36

    A capable and efficient Korean-English open model that you will have to host yourself, because nobody sells it by the token.

    Best for: Korean companies with their own GPUs that want an open model for Korean and English documents and agent work.

    Price: Free weights under the Upstage Solar License (Apache 2.0 plus Solar branding for derivatives). No per-token API; FlexAI dedicated endpoint $3.50/GPU-hour on 16 H100s ($56.00/hour).

    Read our Solar Open 2 review
    3.7/5
  37. 37

    Clover is the most capable open diffusion language model yet on its makers' figures, but as a preview with modest speed gains and no outside tests it is for experimenters.

    Best for: Researchers and inference engineers evaluating diffusion decoding on a modern MoE model before it reaches production quality.

    Price: Open weights under Apache 2.0; no hosted API.

    Read our Clover-35B-A3B-Preview review
    3.6/5
  38. 38

    Still a capable Apache 2.0 workhorse, but Alibaba now sells the same model hosted for less than any third party charges to run the open weights.

    Best for: Teams that need the largest Apache 2.0 Qwen on their own hardware and already have a multi-GPU server or a 256 GB machine to run it.

    Price: Free weights under Apache 2.0; Qwen3.5-Plus on Alibaba Cloud and third-party hosting billed separately.

    Read our Qwen3.5-397B-A17B review
    3.6/5
  39. 39

    A sensible Japanese-tuned Kimi K2.6 with tools built in, but it costs more than newer open models that already handle Japanese well.

    Best for: Japanese companies that want a domestic vendor and a Japanese-tuned model with tools built in, more than the lowest price per token.

    Price: Pay-as-you-go: $0.95 input / $4.00 output per 1M tokens, cached input $0.15; web search $7.00 per 1,000 calls; code execution $0.12 an hour. Free in Sakana Chat (Japan).

    Read our Sakana Namazu review
    3.6/5
  40. 40

    The open model to reach for in Hindi, Tamil or Bengali, as long as nobody asks it to write code or check its own facts.

    Best for: Indian businesses and public bodies that need an open, affordable model fluent in Indian languages.

    Price: Free weights under Apache 2.0. Sarvam API: ₹15 input / ₹5 cached / ₹60 output per 1M tokens; ₹100 free credit on sign-up.

    Read our Sarvam-105B review
    3.6/5
  41. 41

    Xing4.0 is proof that China can train a useful agent model without NVIDIA, and a reasonable cheap coder, but not a reason to leave Qwen.

    Best for: Teams on Huawei Ascend hardware, or those wanting a small Apache-licensed agent model to fine-tune for domain tasks.

    Price: Free weights under Apache 2.0. No first-party API; OpenRouter lists a free endpoint (hosted by Phala) under its free-model request caps.

    Read our Xing4.0-29B-A4B review
    3.6/5
  42. 42

    A properly open Korean frontier model with excellent long-document recall, but by LG's own tables it trails the Chinese open models on most work.

    Best for: Korean organisations that need a permissively licensed domestic model for long Korean documents and can host it themselves.

    Price: Free weights under Apache 2.0. No per-token price published; FriendliAI offers dedicated endpoints billed for GPU time.

    Read our K-EXAONE 2.0 review
    3.4/5
  43. 43

    Maverick was a fair open model in spring 2025, launched under a leaderboard cloud, and nothing Meta has done since gives a reason to start a project on it.

    Best for: Teams with an existing Maverick deployment or a host contract who want cheap multilingual text and image handling and are outside the EU.

    Price: Free weights under the Llama 4 Community License (gated). Hosted per 1M tokens: DigitalOcean $0.188 input / $0.652 output; Parasail $0.35 / $1.00; Google Vertex AI $0.35 / $1.15.

    Read our Llama 4 Maverick review
    3.3/5
  44. 44

    Scout is still one of the cheapest vision-capable models to rent, but its headline context window was never real and Meta itself has left the Llama line behind.

    Best for: Teams already running Llama 4 who need cheap hosted text and image handling outside the EU and can live with an unsupported model line.

    Price: Free weights under the Llama 4 Community License (gated). Hosted: DeepInfra $0.10 input / $0.30 output per 1M tokens ($0.08 / $0.24 on its flex tier); Google Vertex AI $0.25 / $0.70.

    Read our Llama 4 Scout review
    3.2/5

Sources

The main official source behind each ranked review; every review lists its full set.

More guides