Best LLMs for Reasoning and Research Agents
Multi-step planning, tool use and long outputs on hard problems.
Typical request this page prices
- Quality floor: Top 10% of priced models by Intelligence Index (≥ 39.9).
- Costs use each provider’s current standard rates, with cached tokens at the cached-input rate where one is published. Speed is from Artificial Analysis where measured.
Lowest cost that meets the bar
Models that clear the quality floor, ordered by estimated cost for the request above.
| Model | Intelligence | Speed | Cheapest host | Input / output per 1M | Per 1,000 requests |
|---|---|---|---|---|---|
| GLM-5.3-Flash Z AI | 41.8 | 59tok/s | Pareto Inference 17 providers | $0.060 / $0.200 cached $0.012 | $3.47 |
| Claude Haiku 5.5 Anthropic | 43.4 | 238tok/s | DigitalOcean 3 providers | $0.100 / $0.500 cached $0.010 | $6.76 |
| MiMo-V2.6-Pro Xiaomi | 46.3 | 37tok/s | Deep Infra 6 providers | $0.430 / $0.870 cached $0.0036 | $17.41 |
| Grok 4.6 SpaceXAI | 44.3 | - | Velokey 5 providers | $0.800 / $2.40 cached $0.080 | $41.28 |
| GLM-5.3 Z AI | 44.8 | 65tok/s | OpenRouter 16 providers | $0.079 / $4.80 cached $0.059 | $42.42 |
| Step 5 Preview StepFun | 43.7 | 89tok/s | Empirio Labs 2 providers | $1.00 / $2.70 cached $0.050 | $47.40 |
| Gemini 3.8 Flash Google | 40.9 | 138tok/s | GMI Cloud 4 providers | $0.750 / $3.75 cached $0.075 | $50.70 |
| Muse Spark 1.3 Meta | 48.1 | 115tok/s | OpenRouter 2 providers | $1.25 / $4.25 cached $0.150 | $69.40 |
Highest-scoring options
Ordered by Artificial Analysis Intelligence Index, with the cost of the same request.
| Model | Intelligence | Speed | Cheapest host | Input / output per 1M | Per 1,000 requests |
|---|---|---|---|---|---|
| Claude Opus 5.5 Anthropic | 57.6 | 97tok/s | Deep Infra 6 providers | $4.00 / $20.00 cached $0.200 | $263.20 |
| Claude Sonnet 5.5 Anthropic | 56 | 135tok/s | Anthropic 6 providers | $2.00 / $10.00 cached $0.100 | $131.60 |
| Claude Fable 5.1 Anthropic | 53.4 | 72tok/s | DigitalOcean 4 providers | $10.00 / $50.00 cached $0.250 | $649.00 |
| GPT-6 Astra OpenAI | 52.7 | 63tok/s | Velokey 7 providers | $3.00 / $15.00 cached $0.300 | $202.80 |
| GPT-6.1 Sol OpenAI | 51.8 | 65tok/s | Microsoft Azure 6 providers | $2.00 / $10.00 cached $0.100 | $131.60 |
| Claude Opus 5 Anthropic | 50.8 | - | Velokey 7 providers | $2.00 / $10.00 cached $0.200 | $135.20 |
| Claude Fable 5 Anthropic · 1.0M context | 49.6 | - | Velokey 6 providers | $4.00 / $20.00 cached $0.400 | $270.40 |
| Muse Spark 1.3 Meta | 48.1 | 115tok/s | OpenRouter 2 providers | $1.25 / $4.25 cached $0.150 | $69.40 |
Compare every model and provider on the LLM API pricing table.
How to choose an LLM for reasoning & research agents
Research and analysis agents plan, call tools, read results and iterate before writing a long answer. Each step re-sends a growing context, and reasoning models generate thinking tokens that are billed as output. Of the workloads on this site, this has the largest tokens per task in both directions.
Only the top tenth of priced models by intelligence index are compared here. For these tasks a weaker model often fails outright, so a cheaper model is not a saving if it cannot complete the task.
Output rate matters more here than in most workloads because thinking tokens can exceed the visible answer. Compare output speed too: a slow model on an eight-thousand-token answer is a noticeable wait.
Frequently Asked Questions
Are thinking tokens billed?
Yes. Reasoning tokens are billed at the output rate by the providers that publish them, even when they are not returned in the response. Set a reasoning budget where the API allows it to keep costs predictable.
Why is the quality floor so high?
Multi-step tasks compound errors: a model that is right most of the time on single steps can still fail most long tasks. The floor keeps the comparison to models that are realistic candidates.
Can I mix models within one agent?
Yes, and it is common. A top-scoring model plans and checks the work while a cheaper model reads and condenses tool output. Use the cost list on this page to pick the cheaper half of that pairing.