DeepSeek V4
Flash-first production routing and OpenClaw adaptation
V4 Flash as the primary route; V4 Pro only for hard escalation
Flash $0.14 input, $0.028 cache hit, $0.28 output per 1M tokens
Model comparison
DeepSeek V4 vs GPT, Claude, Gemini, and Grok
The comparison is organized by model family. DeepSeek is the V4 Flash and OpenClaw route, GPT is the broad frontier baseline, Claude is the enterprise review route, Gemini is the multimodal/Google route, and Grok is the realtime/xAI route.
Decision matrix
One row per model family, Flash first
This removes duplicate cheapest-row cards while making DeepSeek V4 Flash the primary DeepSeek route and OpenClaw adaptation target.
GPT
General frontier baseline and GPT-compatible customer requirements
GPT-5.4
$2.50 input, $15 output per 1M tokens
Claude
Enterprise review, writing, and code-agent workflows
Claude Sonnet for default Anthropic routing; Opus for top-end review
Sonnet $3/$15; Opus $5/$25 per 1M input/output
Gemini
Multimodal analysis, Google ecosystem work, and long-context checks
Gemini 3.1 Pro Preview
$2 input, $12 output per 1M tokens in the standard tier
Grok
Realtime, agentic, and xAI ecosystem workflows
Grok 4.20
$2 input, $6 output per 1M tokens as a current public reference
Use cases
Practical routing recommendations
Turn the comparison table into production routing rules: default model, upgrade model, and audit model should have separate jobs.
OpenClaw default agent traffic
DeepSeek V4 Flash
At $0.14/M input and $0.28/M output, Flash is the route to adapt for OpenClaw's routine planning, summaries, search, and tool-call narration.
Hard reasoning and code review
DeepSeek V4 Pro + Claude/GPT audit
Use V4 Pro for low-cost hard tasks, then sample with Claude Opus or GPT-5.4 when quality risk is high.
Multimodal product analysis
Gemini 3.1 Pro Preview
Gemini remains the stronger comparison point when image, video, and Google ecosystem features matter.
Realtime and agentic workflows
Grok 4.20
Grok belongs in the comparison set when teams care about realtime behavior, agentic tool calling, or xAI ecosystem fit.
| Family | Best for | Recommended route | Price signal | Caveat |
|---|---|---|---|---|
| DeepSeek V4 | Flash-first production routing and OpenClaw adaptation | V4 Flash as the primary route; V4 Pro only for hard escalation | Flash $0.14 input, $0.028 cache hit, $0.28 output per 1M tokens | Do not make Pro the headline. Flash is the product story; Pro is an upgrade path. |
| GPT | General frontier baseline and GPT-compatible customer requirements | GPT-5.4 | $2.50 input, $15 output per 1M tokens | Good quality baseline, but not the cheapest default route for high-volume text workloads. |
| Claude | Enterprise review, writing, and code-agent workflows | Claude Sonnet for default Anthropic routing; Opus for top-end review | Sonnet $3/$15; Opus $5/$25 per 1M input/output | Separate Claude's enterprise behavior from DeepSeek's cost advantage when building routing rules. |
| Gemini | Multimodal analysis, Google ecosystem work, and long-context checks | Gemini 3.1 Pro Preview | $2 input, $12 output per 1M tokens in the standard tier | Prompt length tiers can change the effective price; check the Google table for >200K prompts. |
| Grok | Realtime, agentic, and xAI ecosystem workflows | Grok 4.20 | $2 input, $6 output per 1M tokens as a current public reference | xAI's pricing table is loaded dynamically; verify account-specific pricing before production use. |