Best AI Models & Chatbots Right Now (Nov 2026)
We may earn a commission if you buy through links on this page. Read the affiliate disclosure · How we review tools
Quick answer. November 2026 picks for reasoning, coding, cheap APIs, open weights, and Chinese models, from official docs and benchmarks.
Last updated: Nov 17, 2026.
For the hardest work this month, the documented frontier pair is Gemini 4 Carbon and Claude Fable 5.2. For everyday coding agents, Gemini 4 Hydrogen is the model with a published price, a free API tier, and a large jump against Gemini 3.8 Flash on Google’s own Terminal-Bench 4.1 numbers. For the lowest flat token rate we could verify, Muse Ember is $0.10 per million input tokens and $0.50 per million output tokens.
This ranking is based on official documentation, benchmarks and public reports. It is not a hands-on test. No score below was measured by us. The dated list of launches is the AI model release timeline. Full launch write-ups: Gemini 4 Carbon, Claude Fable 5.2, Claude Fable 5.2 vs GPT-6.1 Terra, Gemini 4 Hydrogen, Qwen4-Max, OpenAI Images 2.6, GPT Image 2.6 vs Seedream 5.5, Seedance 2.5 Flash, Seedance 2.5 Flash vs Gemini Omni 1.2 Flash, and Claude Meetings.
How this ranking was made
We read vendor announcement posts and the pricing pages linked from them, on Nov 16, 2026. A model is placed in a category when those pages say what it is for and what it costs. A benchmark is included only when the vendor, or a page the vendor cites with a number, printed it. Different versions of Terminal-Bench are not treated as one ladder. On Nov 17, 2026 we added the image-generation section, the meetings section, and the video-generation section, and we linked the Fable versus Terra comparison. Those sections point at the new articles. They do not change the language-model prices below.
There is no star rating on this page. Our usual score table is an editorial fit score for a product we can send you to. These are lab models, and a 4.6 here would pretend we had run them.
Prices we could verify
Standard rates per million tokens, checked Nov 16, 2026. Introductory and regional rates are noted. Cache-hit prices are not the standard input price.
| Model | Input | Output | Notes |
|---|---|---|---|
| Muse Ember | $0.10 | $0.50 | Cached input $0.004. 1M context. Up to 512K output. API id muse-ember-1.0. |
| DeepSeek V4.1 Flash | $0.15 off-peak, $0.30 peak | $0.60 off-peak, $1.20 peak | Cache-miss input. Cache-hit input is $0.003 off-peak and $0.006 peak. API id deepseek-flash. |
| Hy4 | $0.834 | $2.501 | Cached input $0.042, from Tencent’s research post. |
| Gemini 3.8 Flash | $0.75 | $3.75 | Through Dec 31, 2026. Then $1.50 and $7.50. |
| GPT-6.1 Terra | $1 | $5 | Cached input $0.05. API id gpt-6.1-terra. |
| Qwen4-Max | $1.50 | $4.50 | Singapore list price. Cache-hit input $0.075. Beijing / Global is $1.413 and $4.24. |
| Gemini 4 Hydrogen | $1.50 | $7.50 | Flat across 1M context. Implicit cache $0.075. Free tier exists. |
| Gemini 4 Carbon | $2.00 | $10 | Through Nov 30, 2026. Then $4 and $20. No free tier. |
| Kimi K3.1 Pro | $2.00 | $10 | Cache-hit input $0.20. Highspeed is $4 and $20. |
| Claude Fable 5.2 | $8 | $40 | Fast mode $16 and $80. 1M context. 128K max output. |
GPT-6 Cyber’s Nov 5, 2026 launch post does not list a token price. Images 2.6 has its own image-token card, written up in OpenAI Images 2.6, so it is not in this language-model table. Omni 1.2 Flash is priced per second of video, $0.07 at 720p, which is a different meter.
Best overall for hard reasoning
Gemini 4 Carbon and Claude Fable 5.2 are the two models their companies currently describe as the top of the range, and both have a public price.
Google’s Nov 10, 2026 post calls Carbon its strongest model, for cybersecurity and long-horizon work across knowledge tasks, software, and science. The pricing page calls it the most capable reasoning model and says long-decode generation goes up to 1 million tokens. Google says Carbon ranks number 1 on CursorBench 4.0, Perplexity’s WANDR, and Zapier’s AutomationBench, at more than 2 times lower cost per task than the next best external model. Those are Google’s citations. The post does not print the underlying scores in the text we used, so we are not inventing them. Paid API price is $2 / $10 per million tokens through Nov 30, 2026, then $4 / $20. Consumer access in the Gemini app, Antigravity, and Search AI Mode is for Google AI Ultra subscribers.
Anthropic’s Nov 9, 2026 post positions Fable 5.2, and Mythos 5.2 behind a verification program, for the most daunting technical work. The documented spec is a 1 million-token context, 128K max output, and $8 / $40 per million tokens, or $16 / $80 in fast mode. Anthropic says most coding and knowledge work should start on Claude Opus 5.5, and move up to Fable only when the task needs it. Anthropic’s own benchmark table is in Claude Fable 5.2. The head-to-head with GPT-6.1 Terra, including the $1 / $5 list price and the Artificial Analysis index, is Claude Fable 5.2 vs GPT-6.1 Terra. Those numbers are not folded into a cross-lab winner here.
Use Carbon if you are on Google and want the lower introductory rate before Dec 1. Use Fable 5.2 if you are on Claude’s API, Amazon Bedrock, Google Cloud, or Microsoft Foundry, which Anthropic lists as available. This is not a claim that one beat the other in a test we ran.
Best for coding
Gemini 4 Hydrogen is the coding pick for people who need a model they can actually call this week without a frontier contract. Released Nov 12, 2026, it is on the free and paid Gemini API at $1.50 / $7.50 per million tokens. Google says that at the default high thinking level it is 2.5 times faster than Gemini 3.8 Flash and scores 57 percent on Terminal-Bench 4.1 against 18.2 percent. The detail, including what Google did not publish, is in Gemini 4 Hydrogen.
Keep the other coding numbers in their own tests:
- GPT-6.1 Terra, Oct 26, 2026: OpenAI says 50.3 percent on Terminal-Bench 4.0 at maximum reasoning, versus 21.5 percent for GPT-5.6 Terra. Price is $1 / $5 per million tokens. The comparison with Fable 5.2 is Claude Fable 5.2 vs GPT-6.1 Terra.
- Kimi K3.1 Pro: Moonshot says 65.7 percent on Terminal-Bench 4.0. That is Moonshot’s run, and it is version 4.0, not 4.1.
- Qwen4-Max: Alibaba says 6th on Code Arena and does not print the score. See Qwen4-Max.
- Gemini 4 Carbon: Google does not print a Terminal-Bench score in the Nov 10 post. It does say Orchestrator, switching Carbon and Hydrogen in Antigravity, reached near-Carbon success on Terminal-Bench 4.0 with 24 percent less quota and 19 percent less time.
Hydrogen wins this category on availability and on a documented upgrade over Google’s own previous fast model. It does not win it on a cross-lab table, because that table does not exist in the sources above.
See Cursor →Best fast, cheap API
Muse Ember, released Nov 9, 2026, has the lowest flat input and output rates in the price table: $0.10 and $0.50 per million tokens, cached input $0.004, a 1 million-token context, and up to 512K output tokens. Meta’s announcement says it delivers more intelligence per dollar than Claude Haiku 5.5 or GPT-6 Luna at any reasoning setting it compared, and that it matches Haiku’s best score at about a third of the cost. That comparison is Meta’s. We did not reprice Haiku or Luna for this page.
DeepSeek V4.1 Flash is the other cheap API, on a different meter. A cache miss is $0.15 / $0.60 per million tokens off-peak and twice that at peak. A cache hit on input is $0.003 off-peak. The pricing page gives it a 1 million-token context, a 384K maximum output, and vision. If your prompts repeat, DeepSeek’s cache-hit rate is lower than Muse Ember’s cached input. If they do not, Muse Ember’s flat $0.10 input is lower than DeepSeek’s off-peak cache miss.
Gemini 3.8 Flash at $0.75 / $3.75 through Dec 31, 2026 is the cheap option only inside Google’s lineup. It is several times the Muse Ember list price.
Best open-weight model
Kimi K3.1 Pro is the open-weight model to start with. Moonshot’s post, dated Oct 24, 2026 on the blog index, calls it the most capable 3T-class open-weight model the company has shipped, with a 1 million-token context, and says the weights would be released by Nov 4, 2026. The model card now public at moonshotai/Kimi-K3.1-Pro lists 2.8 trillion total parameters, 104 billion active, text and image input, and the Kimi K3.1 Pro License. API price is $2 / $10 per million tokens.
Two other open-weight notes, with narrower claims:
- ERNIE 5.5 Flash Base, Nov 5, 2026, has weights. It is the pre-training checkpoint, 280 billion total and 14 billion active, not an instruction-tuned chatbot. The post-trained preview’s English note says Apache 2.0 weights are coming soon. The Chinese note says by Nov 25, 2026.
- Hy4, Nov 4, 2026, is described by Tencent as open-weight, with an API price of $0.834 / $2.501. We did not confirm a public weight repository on the research post.
- Qwen4-Max is hosted only. Alibaba says smaller open-weight Qwen4 models are still to come.
Best Chinese model
Qwen4-Max is the pick when you want the newest Chinese flagship API. It was released Nov 16, 2026. Alibaba calls it the most capable Qwen, with 2.4 trillion total parameters, 71 billion active, native vision, and a 1 million-token context. On its own Work360 benchmark it reports a better result than Qwen3.8-Max, 35 percent fewer tokens at High effort, and about 60 percent lower average cost per task. Singapore list price is $1.50 / $4.50. The full regional breakdown is in Qwen4-Max.
Use a different Chinese model when the job is different. Kimi K3.1 Pro is the open-weight one. DeepSeek V4.1 Flash is the low token price. Hy4 is the cheaper published list price versus Qwen4-Max’s standard input rate, at $0.834 / $2.501, with vision and computer use claimed by Tencent. ERNIE 5.5 Flash Preview, Nov 11, 2026, is on Qianfan at 280 billion total and 14 billion active. We are not quoting a yuan price for it, because the Qianfan page we saw did not present a clean USD rate.
Gemini 4 Hydrogen is the non-Chinese model most people will compare with Qwen4-Max this week. Hydrogen’s output rate is higher, at $7.50 versus $4.50 in Singapore. Their benchmark tables do not overlap, so the cheaper output token is not a quality result.
Image generation
OpenAI Images 2.6, also called GPT-Image-2.6, was announced Nov 4, 2026. OpenAI says it is in ChatGPT and Codex across all subscription tiers, and in the API, as gpt-image-2.6-flare and gpt-image-2.6-sunburst. The Standard API token price is the same for both. Image output is $30 per million tokens. GIF settings and the comparison with GPT-Image-2.5 are in that article.
ByteDance Seedream 5.5, dated Nov 9, 2026 on the Seed blog, is compared with Images 2.6 on GPT Image 2.6 vs Seedream 5.5. That page carries the use-case split, the Volcano Ark yuan rates, and the note that ByteDance’s GSB claim names GPT Image 2.5 Sunburst rather than 2.6. We did not generate images with either model. Gemini image models and Midjourney are not scored here.
Video generation
Seedance 2.5 Flash was announced Nov 12, 2026. BytePlus ModelArk lists dreamina-seedance-2-5-flash-261115 at $7.0 per million tokens for 480p and 720p. A 5 second, 16:9 example with no video input is $0.34 at 480p and $0.76 at 720p. Clips run 4 to 30 seconds, with generated audio. 1080p and 4K are not on this id. BytePlus’s near-flagship quality line and its 2x speed line are vendor claims.
Seedance 2.5 Flash vs Gemini Omni 1.2 Flash is the use-case split. Flash for a clip past 15 seconds, up to 30, and for video and audio references on the public BytePlus doc. Gemini Omni 1.2 Flash, GA Nov 5, 2026, model id gemini-omni-1.2-flash, for 1080p and 4K and for Google’s listed $0.07 per second at 720p. One Omni generation is 3 to 15 seconds. Veo 3.1 is the older Google video model and is being phased out. This ranking did not generate video with either model.
Meetings and workplace
Claude Meetings is not a new model id. Anthropic announced it on Nov 10, 2026, in beta, on Team and Enterprise. It joins Google Meet as a visible participant. The same day, Anthropic launched Claude Whiteboards and Claude Sheets, both beta on paid plans, and TechCrunch reported that Claude Data, formerly Claude Dashboards, became generally available. Plan limits, beta pricing, and the Fireflies comparison are in that article.
Fireflies remains the meeting-notes link for Zoom, Microsoft Teams, and teams that are not on the Claude plans Meetings names at launch.
Related on this site
- AI model release timeline for the dates and the official links.
- Gemini 4 Carbon for the Nov 10, 2026 frontier model, /swarm, and the introductory price.
- Claude Fable 5.2 for the Nov 9, 2026 release and Mythos 5.2.
- Claude Fable 5.2 vs GPT-6.1 Terra for quality against the $1 / $5 list price.
- Seedance 2.5 Flash for the Nov 12, 2026 video model.
- Seedance 2.5 Flash vs Gemini Omni 1.2 Flash for clips past 15 seconds against 1080p, 4K, and the listed 720p rate.
- Gemini 4 Hydrogen for the Google coding model.
- Qwen4-Max for the Alibaba flagship.
- OpenAI Images 2.6 for Flare, Sunburst, GIFs, and the image token card.
- GPT Image 2.6 vs Seedream 5.5 for the image-model head-to-head.
- Claude Meetings for the Nov 10, 2026 Google Meet beta.
- Mistral Large 4 for the Oct 6, 2026 European launch. That article has not been re-checked against a weight release for this ranking.
Related AI tools
Models are not the product most people buy. If you need a tool around them:
- Cursor for writing code in an editor.
- Fireflies for meeting notes and transcripts.
- Writesonic for marketing copy.
- ElevenLabs for voice, which none of the text models above replace.
Sources
Checked Nov 16, 2026.
- Gemini 4 Hydrogen: https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-hydrogen/
- Gemini 4 Carbon: https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-4-carbon-now-available/
- Gemini API pricing: https://aistudio.google.com/docs/pricing
- Claude Fable 5.2: https://www.anthropic.com/claude-fable-and-mythos-5-2
- Claude Fable 5.2 specs: https://platform.claude.com/docs/en/models/fable-5-2/overview
- GPT-6.1 Terra: https://openai.com/index/introducing-gpt-6-1-terra/
- Muse Ember: https://research.meta.ai/blog/muse-ember
- Qwen4-Max announcement: https://www.alibabacloud.com/blog/alibaba-unveils-qwen4-max-new-ai-teammate-for-work-creativity-and-discovery_603783
- Qwen pricing: https://www.alibabacloud.com/help/en/model-studio/model-pricing
- Kimi K3.1 Pro: https://www.kimi.ai/blog/kimi-k3-1-pro
- Kimi API pricing: https://platform.kimi.ai/docs/pricing
- Kimi model card: https://huggingface.co/moonshotai/Kimi-K3.1-Pro
- DeepSeek pricing: https://api-docs.deepseek.com/quick_start/pricing/
- Hy4: https://hy.tencent.ai/research/hy4
- ERNIE 5.5 Flash Preview: https://ernie.baidu.com/release/ernie-5-5-flash-preview/
- ERNIE 5.5 Flash Base: https://ernie.baidu.com/release/ernie-5-5-flash-base/
- OpenAI Images 2.6, added Nov 17, 2026: https://openai.com/index/introducing-images-2-6/
- Claude Meetings, added Nov 17, 2026: https://claude.com/resources/articles/claude-meetings
- Seedance 2.5 Flash, added Nov 17, 2026: https://docs.byteplus.com/docs/en/modelark/seedance-2-5.md
- Gemini Omni 1.2 Flash announcement, added Nov 17, 2026: https://blog.google/innovation-and-ai/technology/developers-tools/introducing-gemini-omni-1-2-flash/
- Gemini Omni 1.2 Flash model page, added Nov 17, 2026: https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/gemini/omni-1-2-flash
Last updated: Nov 17, 2026. This page contains affiliate links. Rankings follow the public documents above, not a commission.
Frequently asked questions
What is the best AI model in November 2026?
There is no single winner on a shared public test that we could verify across labs. For the hardest work, the models their makers currently position at the frontier, with published prices, are Gemini 4 Carbon and Claude Fable 5.2. Carbon is $2 per million input tokens and $10 per million output tokens through Nov 30, 2026, then $4 and $20, and it is not on the free tier. Fable 5.2 is $8 and $40, with a 1 million-token context and 128K max output. We did not run either model.
What is the best AI model for coding right now?
For a coding model you can call broadly, Gemini 4 Hydrogen is the practical pick on the documents checked Nov 16, 2026. Google says it scores 57 percent on Terminal-Bench 4.1 against 18.2 percent for Gemini 3.8 Flash, at $1.50 per million input tokens and $7.50 per million output tokens. That is Google's benchmark, not ours. Kimi's Terminal-Bench 4.0 score and OpenAI's GPT-6.1 Terra score are different tests and are not ranked against it here.
What is the cheapest capable API?
Among the priced models we checked, Muse Ember has the lowest flat rate: $0.10 per million input tokens and $0.50 per million output tokens, with a 1 million-token context. DeepSeek V4.1 Flash is the cache-hit bargain, at $0.003 per million cached input tokens off-peak. Cheap is a price fact. It is not a quality ranking.
What is the best open-weight model right now?
Kimi K3.1 Pro is the open-weight flagship we could verify. Moonshot announced it on Oct 24, 2026 and said weights would ship by Nov 4, 2026. The public model card lists 2.8 trillion total parameters, 104 billion active, and a 1 million-token context. ERNIE 5.5 Flash Base weights shipped Nov 5, 2026, and that release is a pre-training checkpoint, not a chat model. Qwen4-Max is not open-weight.
What is the best Chinese AI model in November 2026?
Qwen4-Max, released Nov 16, 2026, is the newest flagship its maker calls the most capable Qwen. Singapore list price is $1.50 per million input tokens and $4.50 per million output tokens. Kimi K3.1 Pro is the open-weight alternative at $2 and $10. DeepSeek V4.1 Flash is the low list price. Those are different jobs, and we did not run a three-way test.
Did you test these models yourself?
No. The language-model ranking is based on official documentation, benchmarks and public reports checked Nov 16, 2026. Image generation, Claude Meetings, Seedance 2.5 Flash, the Seedance versus Gemini Omni 1.2 Flash comparison, and the Fable versus Terra comparison were added Nov 17, 2026 from the same kind of sources. Vendor scores are labeled as vendor scores. Where a number was not on an official page, it is not used.