OpenRouter Cuts GPT-5.6 Sol Price 50% (August 2026)

OpenRouter cut the token price of GPT-5.6 Sol by half on its OpenAI Standard route on August 17. The model page and endpoints API continue to show a 50% discount.

August 22 update: OpenAI has now made a separate direct-API cut to $4 input and $20 output per 1M short-context tokens. OpenRouter’s eligible Standard route moved with that base rate and now shows $2 input, $0.20 cached input, $2.50 cache write, and $10 output. The old-versus-new table above preserves the original August 17 route event; today’s direct rates appear in the live table below.

That distinction matters. Teams using OpenRouter’s default model route can access the lower advertised rate; teams calling OpenAI or another OpenRouter provider directly should not assume the same bill.

What changed

The build-time table above shows the old and new OpenRouter route prices. The same 50% reduction applies to cache writes and the long-context tier above 272,000 input tokens. Web-search charges are separate and did not receive the token discount shown on the route.

OpenRouter currently exposes OpenAI Standard, OpenAI Flex, OpenAI Priority, Azure, and Amazon Bedrock endpoints for the model. Its endpoints API marks each OpenAI service tier with the same 50% discount, while Azure and Amazon Bedrock remain undiscounted. Endpoint availability and routing preferences can therefore change the rate that serves a request.

For the direct model card and context rules, see our live OpenAI pricing page. Compare another frontier route on the Anthropic pricing page, or enter your own token mix in the AI token calculator.

What this means for your bill

Every token-priced Sol workload routed through a discounted OpenAI endpoint should cost half as much before separate tools, web search, credits, taxes, or platform-specific charges. A workload that previously consumed a fixed monthly Sol token budget can now process roughly twice as many equivalent tokens for that budget.

Use the build-time old-versus-new table for the verified Standard-route rates. For a workload-level forecast, enter the same input and output volumes in the calculator above, then halve its direct Standard token subtotal. Keep web search and other separately billed tools outside that calculation.

Output-heavy agents gain the largest absolute dollar savings because Sol’s output rate is much higher than its input rate. Long-context applications also benefit, although prompts above the threshold still use a higher tier than shorter requests.

The cut makes Sol more competitive with mid-tier frontier models without requiring a model migration. Our earlier GPT-5.6 tier analysis remains useful for direct-provider buyers, but its Sol price reflects OpenAI’s list card rather than this OpenRouter promotion.

Who benefits—and who does not

Existing OpenRouter customers benefit most because they can retain the same model ID, tool behavior, context window, and evaluations while lowering token spend. Coding agents, research workflows, and long-running tool loops gain when their selected endpoint receives the discount.

Direct OpenAI customers do not receive OpenRouter’s 50% route promotion, although they now receive OpenAI’s separate August 22 direct-rate cut. Requests forced to Azure, Amazon Bedrock, or another endpoint whose card lacks the discount can still cost more. Procurement teams should therefore record both the model and the serving route in cost reports.

Labs impact

AI Pricing Guru Labs already includes GPT-5.6 Sol in its deterministic 49-task leaderboard and uses OpenRouter as the inference route. The public leaderboard deliberately calculates cost from canonical direct-provider list prices, so every model shares one normalized price basis. It will not apply one router promotion to Sol while leaving other models at direct rates.

The discount changes a possible invoice, not Sol’s accepted answers, accuracy, latency, or token counts, so a paid inference rerun would add no model-quality evidence. A future router-price view would need timestamped endpoint selection, actual billed cost, the discount window, routing policy, and fallback-provider records for every run.

What to do now

If you use OpenRouter, inspect the generation metadata for a sample of production calls and confirm the provider tag. Pin the discounted endpoint only if its uptime, data policy, regional requirements, and service tier fit your workload.

Run a small canary and compare billed prompt, cached, cache-write, and completion amounts with your forecast. Then update budgets and routing alerts. Keep a fallback route, but set a maximum acceptable price so failover does not silently erase the savings.

If you call OpenAI directly, use the new $4/$20 Standard card as your source of truth. Evaluate OpenRouter as a separate purchasing route with its own credits, routing, provider controls, and 50% promotion.

Bottom line

This remains a meaningful 50% GPT-5.6 Sol discount for OpenRouter users. OpenAI’s later direct cut improves the base economics for everyone calling OpenAI, while the router promotion can halve eligible routes again. Verify the actual provider and service tier; the model name alone is not enough to predict the bill.

Sources: OpenRouter’s GPT-5.6 Sol model card, OpenRouter’s endpoints API, OpenAI developer pricing, the Hacker News discussion, and the live AI Pricing Guru dataset. Endpoint discounts and prices re-verified August 18, 2026 at 08:15 UTC.