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 AI reasoning models in 2026

The best AI reasoning models right now are Claude Fable 5.2, Gemini 4 Carbon and Gemini 4 Argon, out of 27 we have reviewed. Each is ranked by our rating, with its price, our verdict and who it suits.

By ReviewRamble editors · Updated Nov 11, 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

    The first Fable that independent testing puts clearly above Opus 5.5, and at medium effort it costs less per task than Opus does at max.

    Best for: Teams whose hardest coding, research or analysis agents still fall short on Opus 5.5, and who can accept 30-day data retention.

    Price: API: $8 per 1M input tokens, $40 per 1M output tokens; cache writes $10 (5 min) or $16 (1 hour); cache reads $0.20; Batch API 50% off; fast mode $16/$80.

    Read our Claude Fable 5.2 review
    4.7/5
  2. 2

    Carbon is now the best model anyone can buy from Google, and at Argon's per-token price it beats Opus 5.5 on Artificial Analysis for the same money; Claude Fable 5.2 is still ahead.

    Best for: Gemini API and Google Cloud teams that want frontier reasoning at Opus 5.5's price or below, especially on finance, legal and long-document work.

    Price: API: $2 input / $10 output per 1M tokens until 30 Nov 2026, then $4 / $20; cached input $0.10 (then $0.20); Batch and Flex half price; Priority $3.60 / $18 (then $7.20 / $36). No free tier. Gemini app: Google AI Ultra, $99.99/month (5x) or $199.99/month (20x).

    Read our Gemini 4 Carbon review
    4.6/5
  3. 3

    Google is back at the frontier with Argon, and its native video and audio input is something the Claude and GPT flagships still cannot match.

    Best for: Teams working with video, audio or very large document collections who want a top-tier reasoning model.

    Price: API: $2 input / $10 output per 1M tokens until 30 Nov 2026, then $4 / $20; Batch and Flex half price; cached input $0.10 (then $0.20). No free tier.

    Read our Gemini 4 Argon review
    4.5/5
  4. 4

    Xiaomi now makes the top-scoring open-weight model, at a fraction of closed-model prices; the catch is that self-hosting it is a 16-GPU job.

    Best for: Teams that want the strongest open-weight model for agents and coding, either hosted cheaply on Xiaomi's API or self-hosted on a GPU cluster.

    Price: Open weights under MIT. Xiaomi API: Pro $0.435 input / $0.87 output per 1M tokens, cached input $0.004; Flash $0.14 / $0.28; Pro UltraSpeed $4.35 / $8.70; Batch API half price.

    Read our MiMo-V2.6-Pro review
    4.5/5
  5. 5

    Anthropic's sharpest vulnerability hunter, though for most teams the verification gate, not the model, decides whether it is any use.

    Best for: Security and life-sciences teams that can pass Anthropic's verification and are blocked by Fable 5.2's cyber or biology safeguards.

    Price: API (verified organisations only): $8 per 1M input tokens, $40 per 1M output tokens; cache writes $10 (5 min) or $16 (1 hour); cache reads $0.20; fast mode $16/$80.

    Read our Claude Mythos 5.2 review
    4.4/5
  6. 6

    Three weeks after launch, Anthropic's own cheaper Opus 5.5 overtook Fable 5.1 on most published benchmarks, which leaves it a specialist's model at a generalist's flagship price.

    Best for: Teams with long scientific or research agent tasks where Opus 5.5 at high effort has been tested and found short.

    Price: API: $10 per 1M input tokens, $50 per 1M output tokens; cache writes $12.50 (5 min) or $20 (1 hour); cache reads $0.25; Batch API 50% off.

    Read our Claude Fable 5.1 review
    4.3/5
  7. 7

    DeepSeek-V4-Pro is frontier-class reasoning under an MIT licence at about a quarter of Opus 5.5's per-token price, provided your data can go to China or you own a rack of H200s.

    Best for: Teams that want frontier-class open weights for coding and reasoning, either on their own eight-GPU server or through a very cheap API for non-sensitive work.

    Price: Free MIT-licensed weights. DeepSeek API: $0.66 input / $1.98 output per 1M tokens off-peak, $1.32 / $3.96 at peak; cache hits from $0.022.

    Read our DeepSeek-V4-Pro review
    4.3/5
  8. 8

    A real step up from GPT-5.6-Cyber on exploit work and on staying in scope, at twice GPT-6 Astra's price and behind the strictest vetting OpenAI runs.

    Best for: Established security teams, penetration-testing firms and software vendors with Daybreak Red approval who validate exploits at volume.

    Price: OpenAI API (Daybreak Red only): $20 input / $100 output per 1M tokens, cached input $2; cache writes $25 (30 minutes) or $50 (12 hours). No long-context or Batch price listed.

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

    Kimi K3 tops Moonshot's own agent scorecard, but at $2 per Artificial Analysis task it costs seven times GLM-5.3-Flash for two more index points.

    Best for: Teams building browsing and research agents who want open weights and can self-host or pay frontier-like API prices.

    Price: Open weights under a modified MIT licence. Kimi API: $3 input / $15 output per 1M tokens; cached input $0.30.

    Read our Kimi K3 review
    4.3/5
  10. 10

    A record-setting maths and science model that costs four and a half times as much per task as a sibling scoring one point less.

    Best for: Research, maths and science teams that need the strongest reasoning OpenAI offers and can justify the price.

    Price: API: $10 per 1M input tokens, $50 output; cached input $1; Batch/Flex half price; prompts over 272K input tokens cost 2x input and 1.5x output.

    Read our GPT-6 Astra review
    4.2/5
  11. 11

    Artificial Analysis's harder v5.0 index drops K3.1 Pro from fourth to 22nd, level with GPT-6.1 Sol at medium effort for ten times the cost per task, so its case now rests on the weights due on 4 November.

    Best for: Teams running long terminal and coding agents who can tolerate slow output, and who want an open-weight model to self-host once the weights ship.

    Price: Kimi API: $2 input / $10 output per 1M tokens, cached input $0.20; Highspeed $4 / $20. Batch at 60% of list. A third below Kimi K3. Open weights promised by 4 November 2026.

    Read our Kimi K3.1 Pro review
    4.2/5
  12. 12

    An open flagship that follows instructions and reproduces research better than Claude or GPT on Qwen's numbers, though few will run the weights themselves.

    Best for: Teams doing research, specification-heavy writing and multilingual work through an API who want an open-weights fallback.

    Price: Open weights under the Qwen3.8-Max licence. Qwen Cloud API: $2 input / $6 output per 1M tokens; cache reads $0.17 (explicit) or $0.25 (implicit).

    Read our Qwen3.8-Max review
    4.2/5
  13. 13

    Kimi K3's score at half its cost per task and four GPUs to self-host, but it slips on the agent tests StepFun itself leads with.

    Best for: Teams that want a near-frontier open-weight model for long documents and research that fits on a single GPU server.

    Price: Open weights under the StepFun Model License 1.0. StepFun API: $1.00 input / $2.70 output per 1M tokens, 90% off cached input; same price on OpenRouter and DeepInfra.

    Read our Step 5 Preview review
    4.2/5
  14. 14

    As open as a large model gets, quick and reluctant to make things up, though China's open leaders still out-reason it.

    Best for: Companies that need a large, fast open reasoning model with documented training data, hosted or on their own GPUs.

    Price: Free weights under OpenMDW-1.1. On OpenRouter: DeepInfra $0.40/$1.76 or $0.50/$2.20, Baseten $0.60/$2.40, Venice $0.625/$3.125 per 1M input/output tokens.

    Read our Nemotron 3 Ultra review
    4.1/5
  15. 15

    Fourteen months on, gpt-oss-120b is no longer the strongest open model, but it is still the plainest licence and the widest support for reasoning on one 80GB GPU.

    Best for: Teams that need a permissively licensed reasoning model they can run privately on one GPU, with mature tooling.

    Price: Free weights under Apache 2.0; hosted prices vary by provider.

    Read our gpt-oss-120b review
    4.0/5
  16. 16

    Measurably better than OpenAI's general models on its own life-science tests, but every number is OpenAI's, and access is limited to vetted organisations.

    Best for: Approved pharmaceutical, biotech and academic research teams that want an AI research assistant tuned for biology and chemistry.

    Price: OpenAI API (approved organisations): gpt-rosalind-research $5 input / $25 output per 1M tokens, cached input $0.50; billing from 5 October 2026. US academic labs: up to 30 free Business seats for 12 months, subject to approval.

    Read our GPT-Rosalind review
    4.0/5
  17. 17

    Granite 4.2 is the open family a risk committee will approve fastest, with reasoning that is good for its size and nowhere near the frontier.

    Best for: Companies and developers who want small, permissively licensed reasoning models for local or on-premises agents.

    Price: Free weights under Apache 2.0; watsonx and third-party hosting billed separately.

    Read our Granite 4.2 review
    4.0/5
  18. 18

    Fast and MIT-licensed with near-frontier open-model scores, but it burns tokens thinking and its data terms are not published.

    Best for: Teams that want a fast, permissively licensed open model for latency-sensitive chat and agents, self-hosted or via OpenRouter.

    Price: Open weights under MIT. inclusionAI API: $0.30 input / $0.90 output per 1M tokens, cached $0.06, per Artificial Analysis; OpenRouter lists $0.075 / $0.225.

    Read our Ling-3.1-flash review
    4.0/5
  19. 19

    Kolibri is the strongest case yet for a European-built open model, though the best open model on Aleph Alpha's own German tests comes from Alibaba.

    Best for: German public bodies and regulated firms that need an auditable, European-trained open model for document and retrieval work on their own GPUs.

    Price: Free under Apache 2.0. No public hosted API; enterprise deployment through Aleph Alpha sales, prices not published.

    Read our Kolibri 1 review
    3.9/5
  20. 20

    Sage-A1.5 turns a Qwen base into a credible open research agent with sub-agents built in, though its big wins are on benchmarks few others report.

    Best for: Teams building self-hosted research or browsing agents who want sub-agent fan-out and self-checking in the model rather than in custom harness code.

    Price: Open weights under Apache 2.0; free trial on Shanghai AI Lab's Intern Discovery token plan; paid API prices not published.

    Read our Sage-A1.5 review
    3.9/5
  21. 21

    A clever way to get near-frontier answers without depending on one lab, but you cannot see which models handled your data, and the best scores are Sakana's own.

    Best for: Teams outside the EU that want near-frontier results for multi-step research and coding without depending on a single model provider.

    Price: Fugu Max $2 input / $6 output per 1M tokens ($0.25 cached); Fugu Ultra v2 $5 / $30 ($0.50 cached), $10 / $45 above 272K context; base Fugu at the rate of the top model used. Subscriptions $20, $100 or $200 a month. Web tools $0.007 per call.

    Read our Sakana Fugu (Fugu, Max and Ultra v2) review
    3.9/5
  22. 22

    Beam is the strongest American open model with a genuinely open licence, and its thrift with tokens is real, but the Chinese open models it is pitched against still score well ahead of it.

    Best for: Teams that want a large reasoning model from a US lab under Apache 2.0, especially to distil into smaller models or to run cost-sensitive agents where output tokens dominate the bill.

    Price: Free Apache 2.0 weights. Reflection API free in public beta with daily limits. Hosted: Together AI and Baseten $0.40 in / $0.04 cached / $1.20 out per 1M tokens; OpenRouter from $0.24 / $0.98.

    Read our Beam review
    3.8/5
  23. 23

    GPT-6 Cyber beats it on OpenAI's own exploit and scope tests, but GPT-5.6-Cyber is cheaper and is the cyber model every Daybreak Red customer can get today.

    Best for: Daybreak Red customers not yet approved for GPT-6 Cyber, or running enough exploit validation that the lower price matters more than GPT-6 Cyber's higher scores.

    Price: OpenAI API (Daybreak Red only): $12.50 input / $75 output per 1M tokens, cached input $1.25; cache writes $15.625 (30 minutes) or $31.25 (12 hours). No Batch, Flex or Fast price listed. Amazon Bedrock: $13.75 / $82.50 in US East (Ohio).

    Read our GPT-5.6-Cyber review
    3.8/5
  24. 24

    MAI-Thinking-1 is a credible first model from Microsoft's own lab, priced well for its class, but it trails the models Foundry already sells beside it.

    Best for: Azure teams that want a cheaper reasoning model for maths, coding and agent work under Microsoft's own data terms, and can tolerate preview status.

    Price: Microsoft Foundry: from $2 per 1M input tokens and $8 per 1M output tokens, billed through an Azure subscription; Microsoft AI's own model page publishes no prices.

    Read our MAI-Thinking-1 review
    3.7/5
  25. 25

    Falcon-H1R-7B posts maths scores far above its size, but almost nobody has adopted it and the independent tests are still to come.

    Best for: Teams that need strong competition-style maths reasoning from a model small enough for a laptop, and whose lawyers accept a custom licence.

    Price: Free weights under the Falcon LLM licence (royalty-free; attribution and acceptable use policy required). No public hosted API listed by Artificial Analysis.

    Read our Falcon-H1R-7B review
    3.6/5
  26. 26

    Phi-4-Reasoning-Vision-15B is an MIT-licensed vision model that loses to an 8B Qwen on Microsoft's own scorecard.

    Best for: Teams with a data-centre GPU who need an MIT-licensed English vision model for charts, diagrams and screenshots and are prepared to compare it with Qwen3-VL first.

    Price: Free weights under MIT; Microsoft Foundry hosting billed separately.

    Read our Phi-4-Reasoning-Vision-15B review
    3.4/5
  27. 27

    A credible American open reasoning model for tool-calling agents, overtaken on most tests and now hard to rent.

    Best for: Teams that want a US-trained open reasoning model to self-host or fine-tune for tool-calling agents.

    Price: Free weights under OpenMDW-1.1. Launch API price $0.90 per 1M output tokens on Arcee; no hosted endpoint listed as of 7 November 2026.

    Read our Trinity-Large-Thinking review
    3.4/5

Sources

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

More guides