The short answer: Claude Opus 5.5, GPT-6 Sol and GPT-6 Luna are available through the HopBase API at official list price. Opus 5.5 (claude-opus-5-5) is Anthropic's next-generation Opus: 1M context, 128K max output, $4 input / $20 output per million tokens — 20% below Opus 5 at $5 / $25 — with cache hits at $0.20. GPT-6 Sol (gpt-6-sol) targets complex coding and agent workflows: 1.05M context, $2 / $10, cached input $0.2. GPT-6 Luna (gpt-6-luna) targets high throughput and cost efficiency: 1.05M context, $0.1 / $0.5, cached input $0.01. Above 272K tokens both GPT-6 models switch to a long-context rate (Sol $4 / $15, Luna $0.2 / $0.75). In one line: Opus 5.5 for Claude Code and tool-heavy agents, Sol for Codex and coding agents, Luna for high-volume classification, extraction and routing, and GPT-6 Astra for the hardest problems. Your key's GET /v1/models response is the source of truth for availability.
Published list prices for the three new models
| Model ID | Positioning | Context | Input / output (per 1M tokens) | Cache read |
|---|---|---|---|---|
claude-opus-5-5 |
Anthropic's next-generation Opus | 1M (128K max output) | $4 / $20 | $0.20 |
gpt-6-sol |
Complex coding and agent workflows | 1.05M | $2 / $10 ($4 / $15 above 272K) | $0.2 |
gpt-6-luna |
High throughput, cost efficiency | 1.05M | $0.1 / $0.5 ($0.2 / $0.75 above 272K) | $0.01 |
For reference, GPT-6 Astra is $10 / $50 (cached $1) and Claude Opus 5 is $5 / $25. Opus 5.5 cache writes cost $5 per million for the 5-minute TTL and $8 per million for the 1-hour TTL. These are official list prices; every request's actual charge is itemised in your usage records. Since this post first went out, newer siblings have launched (for example Claude Sonnet 5.5 and GPT-6.1 Sol) — see live pricing at https://hop-base.com/pricing.
What the same request costs on each model
Straight from the list prices above, for one request with 20K input tokens, 2K output tokens and no cache hits:
- GPT-6 Luna: 20K × $0.1/M + 2K × $0.5/M ≈ $0.003
- GPT-6 Sol: 20K × $2/M + 2K × $10/M ≈ $0.06
- Claude Opus 5.5: 20K × $4/M + 2K × $20/M ≈ $0.12
- GPT-6 Astra: 20K × $10/M + 2K × $50/M ≈ $0.30
Luna and Astra are about 100x apart. That is why model choice should follow the task: inside one product, intent detection and field extraction that run hundreds of thousands of times a day belong on Luna, and only the smaller share of requests that need long reasoning chains should go to Sol, Opus 5.5 or Astra. Giving each workload its own API key also separates cost by purpose in your usage records.
How to choose
- Claude Code and tool-heavy agents: Opus 5.5. It is 20% cheaper than Opus 5 and cache hits cost only $0.20, so long sessions that keep re-reading the same system prompt and tool definitions get cheaper the more they hit the cache.
- Codex CLI, coding agents and complex workflows: GPT-6 Sol. At $2 / $10 it offers 1.05M context, enough to put a whole repository in context — just remember the long-context rate beyond 272K.
- High-volume batch work: GPT-6 Luna. Classification, tagging, extraction, rewriting and routing decisions — many calls, each simple — cost a fraction of a cent per call on Luna.
- The hardest problems you can afford: GPT-6 Astra. Run Sol or Opus 5.5 first and escalate only what fails or falls short; that is far cheaper than sending everything to the flagship.
The most reliable method is still to run each candidate on your own evaluation set and compare output quality with the actual charges in your usage records, rather than relying on leaderboards. While you are at it, confirm you are getting the model you pay for with our model fidelity verification method.
Three-minute integration
Claude uses the Anthropic protocol with Base URL https://api.hop-base.com (no /v1); GPT uses the OpenAI-compatible protocol with Base URL https://api.hop-base.com/v1. One key can be enabled for both families; GET /v1/models tells you which IDs it can call.
# First check which models this key can call
curl https://api.hop-base.com/v1/models -H "Authorization: Bearer $HOPBASE_API_KEY"
# Claude Opus 5.5 (Anthropic protocol)
curl https://api.hop-base.com/v1/messages \
-H "x-api-key: $HOPBASE_API_KEY" \
-H "anthropic-version: 2023-06-01" \
-H "Content-Type: application/json" \
-d '{"model":"claude-opus-5-5","max_tokens":64,"messages":[{"role":"user","content":"Reply with exactly: ok"}]}'
# GPT-6 Sol / Luna (OpenAI-compatible; swap the model for gpt-6-luna)
curl https://api.hop-base.com/v1/chat/completions \
-H "Authorization: Bearer $HOPBASE_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model":"gpt-6-sol","messages":[{"role":"user","content":"Reply with exactly: ok"}]}'
For Claude Code, set ANTHROPIC_BASE_URL to https://api.hop-base.com and ANTHROPIC_AUTH_TOKEN to your HopBase key, then switch the model to claude-opus-5-5 (Claude Code guide). For Codex CLI, use the Responses API (wire_api = "responses") with gpt-6-sol as the model (Codex CLI guide).
FAQ
How much cheaper is Opus 5.5 than Opus 5?
At list price, input drops from $5 to $4 and output from $25 to $20 — 20% on both. Cache hits cost $0.20, 0.05x the input price. Opus 5.5 has a 1M context window and 128K max output.
What does 272K mean for GPT-6 Sol and Luna?
Both models have a 1.05M context window, but beyond 272K tokens they are billed at the long-context rate: $4 / $15 for Sol and $0.2 / $0.75 for Luna. Everyday requests under 272K pay the base rates of $2 / $10 and $0.1 / $0.5.
Can my key call these three models?
Availability is scoped per API key. Call GET /v1/models with the key; whatever IDs come back are the ones it can call. Do not guess model names.
Are these the prices I pay?
They are the official list prices published on our pricing page. For enterprise volume, contact us for a quote; every request's charge can be checked line by line in your usage records.
Originally published on the HopBase blog.
Top comments (0)