Which AI model is best for each task?
No single AI model is best at everything. Here is the model that leads each kind of work as of October 2026, the two that follow it, and the published benchmarks behind the ranking. RacLink, every top AI model working together, uses the same idea: it sends each task to the model that's best at it.
Data checked 28 Oct 2026 · every score links to its sourceBest AI model by task
Each model gets a 1–5 rating per task, informed by its published scores. Ties go to the better result on the benchmark that measures that task.
- Best for writing
Gemini 4 Argonthen
Claude Opus 5.5
Claude Fable 5.1
- Best for conversation
Gemini 4 Argonthen
Muse Wonder
Claude Opus 5.5
- Best for coding
Claude Sonnet 5.5then
Qwen3.8 Max
Claude Opus 5.5
- Best for reasoning & math
GPT-6 Astrathen
Kimi K3.1 Pro
Claude Opus 5.5
- Best for web research
Kimi K3.1 Prothen
Claude Opus 5.5
Claude Fable 5.1
- Best for data & analysis
Claude Opus 5.5then
Claude Sonnet 5.5
Kimi K3.1 Pro
- Best for long documents
GPT-6 Astrathen
Gemini 4 Argon
Muse Wonder
- Best for images & PDFs
Gemini 4 Argonthen
Gemini 3.8 Flash
Muse Wonder
- Best for languages
Gemini 4 Argonthen
Gemini 3.8 Flash
Qwen3.8 Max
- Best for quick replies
GPT-6 Lunathen
DeepSeek V4.1 Flash
Gemini 3.8 Flash
Every model compared: what it’s best at, price and context
Price is US dollars per million input / output tokens. Open a model for its use cases, tips and scores.
GPT-6 Astra
OpenAI · Sep 2026- Reasoning & math
- Long documents
Price$10 / $50Context1.1MGood for
- Computer and browser use
- Long-horizon agents
- Scientific reasoning
- Hard math
- Agentic coding
- Deep web research
- Raise reasoning effort for hard problems
- Use fast mode (2x price) for latency
- Cache long prompts: $1 per 1M cached
Scores
- Artificial Analysis Intelligence Index
- 53
- LMArena text Elo
- 1480
- GPQA Diamond
- 96%
- Humanity's Last Exam (with tools)
- 57.2%
- Terminal-Bench 4.0
- 57.9%
Reads Text, Image · Prompts over 272K tokens: 2x input, 1.5x output
RacLink sends coding and reasoning & math hereUse in RacLinkGPT-6 Luna
OpenAI · Sep 2026- Quick replies
Price$0.10 / $0.50Context1.1MGood for
- High-volume automation
- Classification and extraction
- Chat support bots
- Cheap sub-agents
- Summarization
- Use for bulk, cost-sensitive jobs
- Cached input only $0.01/1M
- Escalate hard cases to Astra
Scores
- Artificial Analysis Intelligence Index
- 38
Reads Text, Image · Prompts over 272K tokens: 2x input, 1.5x output
RacLink sends quick replies and conversation hereUse in RacLinkClaude Opus 5.5
Anthropic · Sep 2026- Data & analysis
- Writing
- Web research
Price$4 / $20Context1MGood for
- Agentic coding
- Long-running agents
- Professional knowledge work
- Long-form writing
- Computer use
- Document analysis
- Default pick for most agent workloads
- Use xhigh effort for hardest tasks
- Fast mode: 2x price, faster output
Scores
- Artificial Analysis Intelligence Index
- 58
- LMArena text Elo
- 1508
- Humanity's Last Exam (with tools)
- 67.7%
- Terminal-Bench 4.0
- 66.4%
- ARC-AGI-2
- 93.3%
Reads Text, Image
RacLink sends writing and conversation hereUse in RacLinkClaude Fable 5.1
Anthropic · Sep 2026- Writing
- Web research
Price$10 / $50Context1MGood for
- Multi-hour agent runs
- Scientific research
- Hard reasoning
- Agentic coding
- Complex analysis
- Reserve for the hardest, longest tasks
- Cache reads $0.25/1M cut agent costs
- Try Opus 5.5 first; similar on most work
Scores
- Artificial Analysis Intelligence Index
- 53
- LMArena text Elo
- 1502
- Humanity's Last Exam (with tools)
- 65.6%
- Terminal-Bench 4.0
- 55.8%
- ARC-AGI-2
- 90%
Reads Text, Image
RacLink sends writing and coding hereUse in RacLinkClaude Sonnet 5.5
Anthropic · Sep 2026- Coding
- Data & analysis
Price$2 / $10Context1MGood for
- Agentic coding
- Terminal and CLI agents
- Business writing
- Computer use
- Customer-facing chat
- Best value Claude for daily work
- Medium effort is fast and cheap
- Scores close to Opus 5.5; try it first
Scores
- Artificial Analysis Intelligence Index
- 56
- Humanity's Last Exam (with tools)
- 64.5%
- Terminal-Bench 4.0
- 70.6%
- ARC-AGI-2
- 83.9%
Reads Text, Image
RacLink sends writing and conversation hereUse in RacLinkGemini 4 Argon
Google · Oct 2026- Writing
- Conversation
- Images & PDFs
Price$2 / $10Context1MGood for
- Multimodal analysis
- Agentic coding
- Long documents
- Visual reasoning
- Research synthesis
- Security work
- Price doubles on 2026-12-01
- Strong at video and image inputs
- Try 3.8 Flash first for speed
Scores
- Artificial Analysis Intelligence Index
- 53
- LMArena text Elo
- 1524
- Terminal-Bench 4.0
- 57.4%
- ARC-AGI-2
- 91.3%
Reads Text, Image, Audio, Video · Introductory until 2026-11-30; $4 / $20 from 2026-12-01
RacLink sends writing and conversation hereUse in RacLinkGemini 3.8 Flash
Google · Sep 2026- Images & PDFs
- Languages
- Quick replies
Price$0.75 / $3.75Context1MGood for
- Fast agentic coding
- Multimodal understanding
- Video and audio input
- High-volume apps
- Chart reasoning
- Very fast; good default for agents
- Use high thinking for hard tasks
- Price doubles on 2027-01-01
Scores
- Artificial Analysis Intelligence Index
- 41
- LMArena text Elo
- 1496
- Humanity's Last Exam (no tools)
- 45.4%
- SWE-bench Pro
- 61.6%
- Terminal-Bench 2.1
- 90.8%
Reads Text, Image, Audio, Video · Promotional until 2026-12-31; $1.50 / $7.50 from 2027-01-01
RacLink sends conversation and long documents hereUse in RacLinkGrok 4.7
xAI · Sep 2026- Conversation
- Coding
Price$2 / $6Context500KGood for
- Coding
- Engineering problems
- Professional knowledge work
- Real-time X context
- Cost-efficient reasoning
- Pricing doubles above 200K prompt tokens
- Use xhigh effort on hard tasks
- Fast variant: 2x price, 2x speed
Scores
- Artificial Analysis Intelligence Index
- 46
- Terminal-Bench 4.0
- 38%
- ARC-AGI-2
- 61.4%
Reads Text, Image · Prompts over 200K tokens: $4 / $12
RacLink sends conversation and coding hereUse in RacLinkMuse Wonder
Meta · Oct 2026- Conversation
- Long documents
- Images & PDFs
Price$4 / $20Context1MGood for
- Agentic coding
- Long-running agents
- Legal and professional work
- Video and audio input
- Long-context recall
- Access via Meta Model API
- Cached input only $0.20/1M
- Use max reasoning for hard tasks
Scores
- Artificial Analysis Intelligence Index
- 52
- LMArena text Elo
- 1509
- Terminal-Bench 4.0
- 52%
Reads Text, Image, Audio, Video, Pdf · Muse Wonder Full is only in Muse Code subscriptions, not the API
DeepSeek V4.1 Flash
DeepSeek · Oct 2026Open weights- Quick replies
Price$0.30 / $1.20Context1MGood for
- Budget agentic coding
- High-volume API use
- Math and reasoning
- Self-hosting
- Long documents
- Off-peak hours cost half
- Weights are MIT-licensed
- Called as deepseek-flash (1017 build)
Scores
- Artificial Analysis Intelligence Index
- 45
- LMArena text Elo
- 1475
- Humanity's Last Exam (no tools)
- 42.6%
- Terminal-Bench 4.0
- 42.3%
- ARC-AGI-2
- 78.1%
Reads Text, Image · Peak rate; off-peak (most hours) is $0.15 / $0.60
RacLink sends quick replies and conversation hereUse in RacLinkQwen3.8 Max
Alibaba · Sep 2026Open weights- Coding
- Languages
Price$2 / $6Context1MGood for
- Multi-step coding projects
- Tool orchestration
- Document parsing
- Chart reasoning
- Multilingual tasks
- Good for long autonomous coding runs
- Open weights are text-only, custom licence
- Strong for Chinese and multilingual
Scores
- Artificial Analysis Intelligence Index
- 45
- LMArena text Elo
- 1481
- GPQA Diamond
- 92.6%
- SWE-bench Pro
- 67.7%
- Terminal-Bench 2.1
- 86.6%
Reads Text, Image, Video
RacLink sends coding and long documents hereUse in RacLinkKimi K3.1 Pro
Moonshot AI · Oct 2026- Web research
- Reasoning & math
- Data & analysis
Price$2 / $10Context1MGood for
- Agentic coding
- Terminal and CLI agents
- Deep web research
- Game, 3D and CAD work
- Finance analysis
- Cache hits cost $0.20/1M
- Open weights due by 2026-11-04
- Use max effort for hardest tasks
Scores
- Artificial Analysis Intelligence Index
- 55
- GPQA Diamond
- 94.2%
- Humanity's Last Exam (with tools)
- 70%
- Humanity's Last Exam (no tools)
- 51.1%
- Terminal-Bench 4.0
- 65.7%
Reads Text, Image, Video · Highspeed tier (2-3x faster): $4 / $20
RacLink sends coding and reasoning & math hereUse in RacLinkGLM-5.3
Z.ai · Aug 2026Open weights- Conversation
- Coding
Price$1.40 / $4.40Context1MGood for
- Agentic coding
- Automation workflows
- Security analysis
- Self-hosting
- Budget reasoning
- Reasoning is always on
- Flash variant for cheap bulk work
- Check custom licence before hosting
Scores
- Artificial Analysis Intelligence Index
- 45
- LMArena text Elo
- 1479
Reads Text
Mistral Large 4 (Preview)
Mistral AI · Oct 2026- Languages
- Writing
Price$1.36 / $4.18Context1MGood for
- European data residency
- Agentic coding
- Security analysis
- Image understanding
- Multilingual business
- Preview: behaviour may still change
- Cached input only $0.14/1M
- Good fit for EU compliance needs
Scores
- Artificial Analysis Intelligence Index
- 38
- Terminal-Bench 4.0
- 28.3%
Reads Text, Image · Public preview; weights promised for end of October 2026
RacLink sends languages and writing hereUse in RacLinkCommand A+
Cohere · May 2026Open weights- Languages
- Quick replies
Price—Context128KGood for
- Enterprise RAG
- Private deployment
- Multilingual (48 languages)
- Document processing
- Tool use
- Runs on 2x H100 or 1x B200
- Apache 2.0: fully self-hostable
- Pair with Cohere Rerank for RAG
Scores
- Artificial Analysis Intelligence Index
- 13
Reads Text, Image
AI model benchmarks: Claude vs GPT vs Gemini and more
The best published score in each column is highlighted. A dash means the provider did not publish it. Benchmarks fewer than 3 models report are left out.
| Model | AA Index | TB 4.0 | Arena Elo | ARC-AGI-2 | HLE+tools | GPQA | HLE |
|---|---|---|---|---|---|---|---|
| 58 | 66.4% | 1508 | 93.3% | 67.7% | — | — | |
| 56 | 70.6% | — | 83.9% | 64.5% | — | — | |
| 55 | 65.7% | — | — | 70% | 94.2% | 51.1% | |
| 53 | 57.4% | 1524 | 91.3% | — | — | — | |
| 53 | 55.8% | 1502 | 90% | 65.6% | — | — | |
| 53 | 57.9% | 1480 | 95% | 57.2% | 96% | — | |
| 52 | 52% | 1509 | — | — | — | — | |
| 46 | 38% | — | 61.4% | — | — | — | |
| 45 | — | 1481 | — | — | 92.6% | — | |
| 45 | — | 1479 | — | — | — | — | |
| 45 | 42.3% | 1475 | 78.1% | — | — | 42.6% | |
| 41 | — | 1496 | 89.2% | — | — | 45.4% | |
| 38 | — | — | — | — | — | — | |
| 38 | 28.3% | — | — | — | — | — | |
| 13 | — | — | — | — | — | — |
- AA Index
- Artificial Analysis Intelligence Index: composite of ten independent evals.
- TB 4.0
- Terminal-Bench 4.0: agentic command-line tasks (hard).
- Arena Elo
- LMArena text Elo: human preference head-to-head rating.
- ARC-AGI-2
- ARC-AGI-2: novel visual reasoning puzzles.
- HLE+tools
- Humanity's Last Exam (with tools): hardest expert questions, search and code.
- GPQA
- GPQA Diamond: phD-level science questions.
- HLE
- Humanity's Last Exam (no tools): hardest expert questions, no tools.
The active model sheet: what RacLink routes on
Every model RacLink can run today, with the facts and the starting quality per kind of work that the model router decides on before it has measured a model in your workspace. A quality from a published benchmark is shown plain; one from the model’s tier (no benchmark published for that work) is dimmed. After eight measured runs the workspace’s own record outweighs either.
| Model | Tier | Context | In /1M | Out /1M | Speed | Writing | Conversation | Coding | Reasoning & math | Web research | Data & analysis | Long documents | Languages | Fit for |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Alibaba Cloud | ||||||||||||||
| Qwen3.7 Max | flagship | 1M | $2.5 | $7.5 | slow | 85% | 85% | 85% | 85% | 85% | 85% | 90% | 85% | Code, Documents, Spreadsheets & data, Long documents |
| Qwen3.7 Max Previewpreview | reasoning | 1M | $2.5 | $7.5 | slow | 85% | 85% | 85% | 85% | 85% | 85% | 90% | 85% | Code, Documents, Spreadsheets & data, Long documents |
| Qwen3.8 Max | flagship | 1M | $2 | $6 | slow | 80% | 80% | 89% | 89% | 82% | 85% | 90% | 82% | Code, Documents, Spreadsheets & data, Presentations, Images & PDFs, Long documents |
| Qwen3.8 2.4T-A95B (open-weight) | flagship | 1M | $2 | $6 | slow | 85% | 85% | 85% | 85% | 85% | 85% | 90% | 85% | Code, Documents, Spreadsheets & data, Long documents |
| Qwen3.6 27B (open-weight) | balanced | 256k | $0.6 | $3.6 | medium | 75% | 75% | 75% | 75% | 75% | 75% | 70% | 75% | Code, Documents, Spreadsheets & data |
| Qwen3.5 397B-A17B (open-weight) | balanced | 256k | $0.6 | $3.6 | medium | 75% | 75% | 75% | 75% | 75% | 75% | 70% | 75% | Code, Documents, Spreadsheets & data |
| Qwen3.6 Plus | balanced | 1M | $0.5 | $3 | medium | 75% | 75% | 75% | 75% | 75% | 75% | 90% | 75% | Code, Documents, Spreadsheets & data, Long documents |
| Qwen3.8 27B (open-weight) | fast | 1M | $0.5 | $3 | fast | 60% | 60% | 60% | 60% | 60% | 60% | 90% | 60% | Long documents |
| Qwen3.5 122B-A10B (open-weight) | balanced | 256k | $0.4 | $3.2 | medium | 75% | 75% | 75% | 75% | 75% | 75% | 70% | 75% | Code, Documents, Spreadsheets & data |
| Qwen3.5 Plus | balanced | 1M | $0.4 | $2.4 | medium | 75% | 75% | 75% | 75% | 75% | 75% | 90% | 75% | Code, Documents, Spreadsheets & data, Long documents |
| Qwen3.6 35B-A3B (open-weight) | fast | 256k | $0.375 | $2.25 | fast | 60% | 60% | 60% | 60% | 60% | 60% | 70% | 60% | — |
| Qwen3.5 27B (open-weight) | fast | 256k | $0.3 | $2.4 | fast | 60% | 60% | 60% | 60% | 60% | 60% | 70% | 60% | — |
| Qwen3.7 Plus | balanced | 1M | $0.4 | $1.6 | medium | 75% | 75% | 75% | 75% | 75% | 75% | 90% | 75% | Code, Documents, Spreadsheets & data, Presentations, Images & PDFs, Long documents |
| Qwen3.5 35B-A3B (open-weight) | fast | 256k | $0.25 | $2 | fast | 60% | 60% | 60% | 60% | 60% | 60% | 70% | 60% | — |
| Qwen3 Coder Flash | code | 1M | $0.3 | $1.5 | medium | 80% | 80% | 80% | 80% | 80% | 80% | 90% | 80% | Code, Documents, Spreadsheets & data, Long documents |
| Qwen3.6 Flash | fast | 1M | $0.25 | $1.5 | fast | 60% | 60% | 60% | 60% | 60% | 60% | 90% | 60% | Long documents |
| Qwen3 VL Plus | balanced | 262k | $0.2 | $1.6 | medium | 75% | 75% | 75% | 75% | 75% | 75% | 70% | 75% | Code, Documents, Spreadsheets & data, Presentations, Images & PDFs |
| Qwen3.8 Flash | fast | 1M | $0.15 | $0.47 | fast | 60% | 60% | 60% | 60% | 60% | 60% | 90% | 60% | Long documents |
| Qwen3.5 Flash | fast | 1M | $0.1 | $0.4 | fast | 60% | 60% | 60% | 60% | 60% | 60% | 90% | 60% | Long documents |
| Qwen3.7 Flash | fast | 1M | $0.03 | $0.13 | fast | 60% | 60% | 60% | 60% | 60% | 60% | 90% | 60% | Long documents |
| Anthropic | ||||||||||||||
| Claude Fable 5.1 | reasoning | 1M | $10 | $50 | slow | 85% | 85% | 83% | 89% | 88% | 87% | 90% | 87% | Code, Documents, Spreadsheets & data, Presentations, Images & PDFs, Long documents |
| Claude Opus 5.5 | flagship | 1M | $4 | $20 | slow | 88% | 88% | 88% | 90% | 90% | 90% | 90% | 90% | Code, Documents, Spreadsheets & data, Presentations, Images & PDFs, Long documents |
| Claude Sonnet 5.5 | balanced | 1M | $2 | $10 | medium | 89% | 89% | 90% | 87% | 89% | 89% | 90% | 89% | Code, Documents, Spreadsheets & data, Presentations, Images & PDFs, Long documents |
| Claude Haiku 5.5 | fast | 1M | $0.1 | $0.5 | fast | 60% | 60% | 60% | 60% | 60% | 60% | 90% | 60% | Images & PDFs, Long documents |
| Cohere | ||||||||||||||
| Aya Expanse 32B | balanced | 128k | $0.5 | $1.5 | medium | 75% | 75% | 75% | 75% | 75% | 75% | 55% | 75% | Code, Documents, Spreadsheets & data |
| Command R (08-2024) | fast | 128k | $0.15 | $0.6 | fast | 60% | 60% | 60% | 60% | 60% | 60% | 55% | 60% | — |
| Command R7B | fast | 128k | $0.037 | $0.15 | fast | 60% | 60% | 60% | 60% | 60% | 60% | 55% | 60% | — |
| DeepSeek | ||||||||||||||
| DeepSeek V4 Pro | reasoning | 1M | $1.32 | $3.96 | slow | 85% | 85% | 85% | 85% | 85% | 85% | 90% | 85% | Code, Documents, Spreadsheets & data, Long documents |
| DeepSeek V4.1 Flash | fast | 1M | $0.3 | $1.2 | fast | 79% | 79% | 76% | 86% | 82% | 82% | 90% | 82% | Code, Documents, Spreadsheets & data, Presentations, Images & PDFs, Long documents |
| Gemini 3.1 Pro Previewpreview | reasoning | 1.05M | $2 | $12 | slow | 85% | 85% | 85% | 85% | 85% | 85% | 90% | 85% | Code, Documents, Spreadsheets & data, Presentations, Images & PDFs, Long documents |
| Gemini 3.1 Pro Preview (Custom Tools)preview | code | 1.05M | $2 | $12 | medium | 80% | 80% | 80% | 80% | 80% | 80% | 90% | 80% | Code, Documents, Spreadsheets & data, Presentations, Images & PDFs, Long documents |
| Gemini 4 Argon | flagship | 1.05M | $2 | $10 | slow | 88% | 88% | 83% | 89% | 87% | 87% | 90% | 87% | Code, Documents, Spreadsheets & data, Presentations, Images & PDFs, Long documents |
| Gemini 3.8 Flash | balanced | 1.05M | $0.75 | $3.75 | medium | 81% | 81% | 88% | 89% | 80% | 80% | 90% | 80% | Code, Documents, Spreadsheets & data, Presentations, Images & PDFs, Long documents |
| Gemini 3.6 Flash | balanced | 1.05M | $0.75 | $3.75 | medium | 75% | 75% | 75% | 75% | 75% | 75% | 90% | 75% | Code, Documents, Spreadsheets & data, Presentations, Images & PDFs, Long documents |
| Gemini 3 Flash Previewpreview | balanced | 1.05M | $0.5 | $3 | medium | 75% | 75% | 75% | 75% | 75% | 75% | 90% | 75% | Code, Documents, Spreadsheets & data, Presentations, Images & PDFs, Long documents |
| Gemini 3.5 Flash-Lite | fast | 1.05M | $0.3 | $2.5 | fast | 60% | 60% | 60% | 60% | 60% | 60% | 90% | 60% | Images & PDFs, Long documents |
| Meta | ||||||||||||||
| Muse Wonderpreview | flagship | 1.05M | $4 | $20 | slow | 86% | 86% | 81% | 85% | 86% | 86% | 90% | 86% | Code, Documents, Spreadsheets & data, Presentations, Images & PDFs, Long documents |
| Muse Spark 1.3 | balanced | 1.05M | $1.25 | $4.25 | medium | 75% | 75% | 75% | 75% | 75% | 75% | 90% | 75% | Code, Documents, Spreadsheets & data, Presentations, Images & PDFs, Long documents |
| MiniMax | ||||||||||||||
| MiniMax M3 | flagship | 1M | $0.3 | $1.2 | slow | 85% | 85% | 85% | 85% | 85% | 85% | 90% | 85% | Code, Documents, Spreadsheets & data, Presentations, Images & PDFs, Long documents |
| MiniMax M3.1 Flash | fast | 1M | $0.15 | $0.6 | fast | 60% | 60% | 60% | 60% | 60% | 60% | 90% | 60% | Images & PDFs, Long documents |
| Mistral AI | ||||||||||||||
| Mistral Medium 3.5 | flagship | 256k | $1.5 | $7.5 | slow | 85% | 85% | 85% | 85% | 85% | 85% | 70% | 85% | Code, Documents, Spreadsheets & data, Presentations, Images & PDFs |
| Mistral Large 4preview | flagship | 1M | $1.36 | $4.18 | slow | 78% | 78% | 69% | 85% | 78% | 78% | 90% | 78% | Documents, Spreadsheets & data, Presentations, Images & PDFs, Long documents |
| Mistral Large 3 | balanced | 256k | $0.5 | $1.5 | medium | 75% | 75% | 75% | 75% | 75% | 75% | 70% | 75% | Code, Documents, Spreadsheets & data, Presentations, Images & PDFs |
| Codestral 25.08 | code | 128k | $0.3 | $0.9 | medium | 80% | 80% | 80% | 80% | 80% | 80% | 55% | 80% | Code, Documents, Spreadsheets & data |
| Mistral Small 4 | fast | 256k | $0.15 | $0.6 | fast | 60% | 60% | 60% | 60% | 60% | 60% | 70% | 60% | Images & PDFs |
| Ministral 3 14B | fast | 256k | $0.2 | $0.2 | fast | 60% | 60% | 60% | 60% | 60% | 60% | 70% | 60% | Images & PDFs |
| Ministral 3 8B | fast | 256k | $0.15 | $0.15 | fast | 60% | 60% | 60% | 60% | 60% | 60% | 70% | 60% | Images & PDFs |
| Ministral 3 3B | fast | 256k | $0.1 | $0.1 | fast | 60% | 60% | 60% | 60% | 60% | 60% | 70% | 60% | Images & PDFs |
| Moonshot AI | ||||||||||||||
| Kimi K3 | flagship | 1.05M | $3 | $15 | slow | 85% | 85% | 85% | 85% | 85% | 85% | 90% | 85% | Code, Documents, Spreadsheets & data, Presentations, Images & PDFs, Long documents |
| Kimi K2.7 Code HighSpeed | code | 262k | $1.9 | $8 | medium | 80% | 80% | 80% | 80% | 80% | 80% | 70% | 80% | Code, Documents, Spreadsheets & data, Presentations, Images & PDFs |
| Kimi K2.7 Code | code | 262k | $0.95 | $4 | medium | 80% | 80% | 80% | 80% | 80% | 80% | 70% | 80% | Code, Documents, Spreadsheets & data, Presentations, Images & PDFs |
| Kimi K2.6 | balanced | 262k | $0.95 | $4 | medium | 75% | 75% | 75% | 75% | 75% | 75% | 70% | 75% | Code, Documents, Spreadsheets & data, Presentations, Images & PDFs |
| OpenAI | ||||||||||||||
| GPT-6 Astra | flagship | 1.05M | $10 | $50 | slow | 82% | 82% | 84% | 88% | 86% | 88% | 90% | 87% | Code, Documents, Spreadsheets & data, Presentations, Images & PDFs, Long documents |
| GPT-6.1 Sol | balanced | 1.05M | $2 | $10 | medium | 75% | 75% | 75% | 75% | 75% | 75% | 90% | 75% | Code, Documents, Spreadsheets & data, Presentations, Images & PDFs, Long documents |
| GPT-6 Sol | code | 1.05M | $2 | $10 | medium | 80% | 80% | 80% | 80% | 80% | 80% | 90% | 80% | Code, Documents, Spreadsheets & data, Presentations, Images & PDFs, Long documents |
| GPT-6 Luna | fast | 1.05M | $0.1 | $0.5 | fast | 78% | 78% | 60% | 60% | 78% | 78% | 90% | 78% | Documents, Presentations, Images & PDFs, Long documents |
| Perplexity | ||||||||||||||
| Sonar (Agent API) | fast | 128k | $0.25 | $2.5 | fast | 60% | 60% | 60% | 60% | 60% | 60% | 55% | 60% | — |
| xAI | ||||||||||||||
| Grok 4.7 | flagship | 500k | $2 | $6 | slow | 83% | 83% | 74% | 78% | 83% | 83% | 80% | 83% | Documents, Spreadsheets & data, Presentations, Images & PDFs, Long documents |
| Grok 4.6 | flagship | 500k | $2 | $6 | slow | 85% | 85% | 85% | 85% | 85% | 85% | 80% | 85% | Code, Documents, Spreadsheets & data, Presentations, Images & PDFs, Long documents |
| Grok 4.5 | code | 500k | $2 | $6 | medium | 80% | 80% | 80% | 80% | 80% | 80% | 80% | 80% | Code, Documents, Spreadsheets & data, Presentations, Images & PDFs, Long documents |
| Grok 4.3 | balanced | 1M | $1.25 | $2.5 | medium | 75% | 75% | 75% | 75% | 75% | 75% | 90% | 75% | Code, Documents, Spreadsheets & data, Presentations, Images & PDFs, Long documents |
| Grok 4.20 (Reasoning) | reasoning | 1M | $1.25 | $2.5 | slow | 85% | 85% | 85% | 85% | 85% | 85% | 90% | 85% | Code, Documents, Spreadsheets & data, Presentations, Images & PDFs, Long documents |
| Grok 4.20 (Non-Reasoning) | fast | 1M | $1.25 | $2.5 | fast | 60% | 60% | 60% | 60% | 60% | 60% | 90% | 60% | Images & PDFs, Long documents |
| Grok Build 0.1 | code | 256k | $1 | $2 | medium | 80% | 80% | 80% | 80% | 80% | 80% | 70% | 80% | Code, Documents, Spreadsheets & data, Presentations, Images & PDFs |
| Z.ai | ||||||||||||||
| GLM-5.3 | flagship | 1M | $1.4 | $4.4 | slow | 80% | 80% | 85% | 85% | 82% | 82% | 90% | 82% | Code, Documents, Spreadsheets & data, Long documents |
| GLM-5.2 | flagship | 1M | $1.4 | $4.4 | slow | 85% | 85% | 85% | 85% | 85% | 85% | 90% | 85% | Code, Documents, Spreadsheets & data, Long documents |
| GLM-5.3-FlashX | fast | 1M | $0.37 | $1.25 | fast | 60% | 60% | 60% | 60% | 60% | 60% | 90% | 60% | Images & PDFs, Long documents |
| GLM-5.3-Flash | fast | 1M | $0.15 | $0.5 | fast | 60% | 60% | 60% | 60% | 60% | 60% | 90% | 60% | Images & PDFs, Long documents |
Prices are each company’s published card (long-context tiers bill higher; see the catalog). Context is the window the model reads; the router never gives a prompt to a model whose window is smaller. “Fit for” is read off the qualities and the catalog: code (strong at coding or a code model), documents (strong at writing), spreadsheets (strong at data and reasoning), presentations (strong at writing and reads images), images and PDFs (reads them), long documents (a window of 400k tokens or more).
How do I use more than one of these models?
OpenAI, Anthropic, Google, xAI, DeepSeek, Alibaba, Moonshot AI and Mistral AI are built into RacLink, so there are no API keys to set up. Each task goes to the model that fits it: a strong model for hard reasoning, a fast and cheap one for quick replies. AI agents then do the work in your tools, with your approval before anything goes out.
Prefer to stay in Claude or ChatGPT? Connect Claude to your CRM and business data with the RacLink MCP server. Plans are a flat price for the whole workspace with unlimited members; see RacLink pricing.
Questions about choosing an AI model
- Which AI model is best overall?
- No single AI model is best at everything. On the Artificial Analysis Intelligence Index, the highest score in this comparison is 58, held by Claude Opus 5.5. On LMArena's human-preference ranking, the top rating is 1524 Elo, held by Gemini 4 Argon. The best choice depends on the task, which is why this page ranks models task by task.
- Which AI model is best for coding?
- For coding, the ranking here is Claude Sonnet 5.5, then Qwen3.8 Max and Claude Opus 5.5. Ties are broken on agentic coding benchmarks (Terminal-Bench and SWE-Bench Pro) where the provider published a score.
- Which AI model is best for writing?
- For writing, the ranking here is Gemini 4 Argon, then Claude Opus 5.5 and Claude Fable 5.1. Ties are broken on LMArena, where people vote for the answer they prefer.
- Claude vs GPT vs Gemini: which one should I use?
- Anthropic's Claude models rank first for coding and data & analysis. OpenAI's GPT models rank first for reasoning & math, long documents and quick replies. Google's Gemini models rank first for writing, conversation, images & PDFs and languages. You don't have to choose just one: RacLink sends each task to the model that is best at it.
- How up to date is this comparison?
- The models and benchmark scores were last checked on 28 October 2026. Every score links to its source, and a benchmark a provider did not publish is shown as a dash, never an estimate.
- Can I use these AI models in RacLink?
- Yes. OpenAI, Anthropic, Google, xAI, DeepSeek, Alibaba, Moonshot AI and Mistral AI are built into RacLink, so there are no API keys to set up. Pick a model yourself, or let RacLink send each task to the model that fits it. If you already pay for a provider, you can add your own key too. You can also connect Claude and other AI apps to your RacLink workspace through the RacLink MCP server.
Sources for every score
- blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-4-argon/
- deepmind.google/models/model-cards/gemini-4-argon/
- ai.google.dev/gemini-api/docs/latest-model
- ai.google.dev/gemini-api/docs/pricing
- arcprize.org/results/google-gemini-4-argon
- artificialanalysis.ai/leaderboards/models
- arena.ai/leaderboard/text
- research.meta.ai/blog/muse-wonder
- research.meta.ai/static/muse-wonder-model-card
- openrouter.ai/meta/muse-wonder
- techcrunch.com/2026/10/12/meta-releases-its-first-frontier-model-muse-wonder/
- venturebeat.com/technology/metas-new-muse-wonder-and-muse-wonder-full-models-achieve-frontier-performance-but-only-the-weaker-one-is-available-via-api
- artificialanalysis.ai/leaderboards/models
- arena.ai/leaderboard/text