GPT-5.6 is OpenAI’s newest announced model family, but it is not a generally available product lane. Sol, Terra, and Luna are available only to selected API organizations and Codex workspaces during the preview. There is no public application or waitlist, and ChatGPT does not include the family yet.
Use this page for model selection and pricing. Use the Sol preview investigation for the system card’s coding-agent risks and the frontier access guide for policy context.
Family and Pricing
Prices are per 1 million tokens. Cache-read prices apply the published 90% discount; cache writes cost 1.25 times normal input.
| Model | Position | Input | Cache read / write | Output | Current access |
|---|---|---|---|---|---|
| GPT-5.6 Sol | Flagship | $5.00 | $0.50 / $6.25 | $30.00 | Selected API organizations and Codex workspaces |
| GPT-5.6 Terra | Balanced | $2.50 | $0.25 / $3.125 | $15.00 | Same restricted preview |
| GPT-5.6 Luna | Fast, lowest cost | $1.00 | $0.10 / $1.25 | $6.00 | Same restricted preview |
OpenAI also documents explicit cache breakpoints and a 30-minute minimum cache life. Context windows, public rate limits, and a general-availability date have not been published. Do not copy GPT-5.5 limits or context specifications into GPT-5.6 rows.
Which Tier Would Fit?
| Need | Preview tier to evaluate | Current usable fallback |
|---|---|---|
| Hard coding, science, or cyber evaluation | Sol | GPT-5.5 or Claude Opus 4.8 |
| Balanced capability and price | Terra | GPT-5.5, Claude Sonnet 4.6, or GLM-5.2 |
| Lowest GPT-5.6 token price and latency | Luna | Kimi K2.7 Code, GLM-5.2, or another generally available value lane |
These are evaluation hypotheses, not production rankings. Terra’s claimed GPT-5.5 competitiveness and Luna’s speed position come from OpenAI. AIHackers has not run a controlled repository evaluation for any GPT-5.6 tier.
Benchmark Evidence
benchmark artifact
GPT-5.6 Preview Evidence and Active Baselines
| Model | Provider | Status | Context | Input price | Output price | Coding signal | Tool-use signal | Benchmark evidence | Speed | Verdict | Sources | Checked |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| GPT-5.6 Sol | OpenAI | preview Selected API organizations and Codex workspaces only; no public enrollment and no ChatGPT access during preview. | not published | $5.00 / 1M | $30.00 / 1M | OpenAI reports a new state of the art on Terminal-Bench 2.1; expanded and independent results are pending. | API and Codex preview; max reasoning and ultra multi-agent modes are vendor-documented. |
| OpenAI announced a selected-customer Cerebras preview for July; production latency is not verified. | Restricted-preview evaluation only; keep GPT-5.5 as the active OpenAI baseline. | OpenAI GPT-5.6 launch [archive], OpenAI GPT-5.6 preview access [archive], OpenAI GPT-5.6 Preview system card [archive] | 2026-06-28 |
| GPT-5.6 Terra | OpenAI | preview Same selected-organization API/Codex preview gate as Sol; not in ChatGPT. | not published | $2.50 / 1M | $15.00 / 1M | OpenAI positions Terra as competitive with GPT-5.5; no independent normalized result is verified. | API and Codex preview only. |
| not verified | Restricted-preview balanced lane; listed pricing does not make it a generally usable value route. | OpenAI GPT-5.6 launch [archive], OpenAI GPT-5.6 preview access [archive], OpenAI GPT-5.6 Preview system card [archive] | 2026-06-28 |
| GPT-5.6 Luna | OpenAI | preview Same selected-organization API/Codex preview gate as Sol; not in ChatGPT. | not published | $1.00 / 1M | $6.00 / 1M | OpenAI positions Luna as the fastest and lowest-cost GPT-5.6 tier; independent coding quality is not verified. | API and Codex preview only. |
| Vendor-positioned as fastest; measured production latency is not verified. | Restricted-preview low-cost tier; do not present it as a public budget default. | OpenAI GPT-5.6 launch [archive], OpenAI GPT-5.6 preview access [archive], OpenAI GPT-5.6 Preview system card [archive] | 2026-06-28 |
| GPT-5.5 | OpenAI | active Generally available OpenAI baseline while GPT-5.6 remains in limited preview. | 1.05M API; 400K Codex | $5.00 / 1M | $30.00 / 1M | not verified | not verified | not verified | not verified | Primary coding seat while ChatGPT/Codex limits fit the workload. | OpenAI GPT-5.5 API model page, OpenAI GPT-5.5 ChatGPT limits, Artificial Analysis: GPT-5.5, LMArena leaderboard dataset | 2026-06-28 |
| Claude Opus 4.8 | Anthropic | active Current generally available Opus-tier premium baseline. | 1M | $5.00 / 1M | $25.00 / 1M | Practical premium Claude baseline; AIHackers recommends task-level comparison rather than blanket default routing. | Claude API and Claude-native workflow baseline; third-party routing must follow Anthropic terms. |
| Artificial Analysis measured 60.4 output tokens/s; provider and workload latency vary. | Premium review, architecture, hard-debugging, and final-arbitration lane. | Claude models overview [archive], Claude API pricing [archive], Artificial Analysis: Claude Opus 4.8 [archive], Artificial Analysis Intelligence Index v4.1, LMArena leaderboard dataset, Berkeley Function Calling Leaderboard | 2026-06-28 |
| GLM-5.2 | Z.AI | active Current Z.AI flagship coding model and supported-tool value lane. | 1M | $1.40 / 1M | $4.40 / 1M | Z.AI reports 62.1 on SWE-Bench Pro and 81.0 on Terminal-Bench 2.1. | Supported-tool coding lane; BFCL score not imported. |
| Artificial Analysis flags higher output-token use; measure total cost per successful task. | July value pick to test for supported coding-tool workflows; keep Opus/GPT for final arbitration until local evals pass. | Z.AI GLM-5.2 overview [archive], Z.AI pricing [archive], Artificial Analysis: GLM-5.2 article [archive], Artificial Analysis Intelligence Index v4.1, SWE-bench, Berkeley Function Calling Leaderboard | 2026-06-28 |
| Kimi K2.7 Code | Moonshot AI | active Current Kimi coding API release; HighSpeed is the same model at higher token prices. | 256K | $0.95 / 1M | $4.00 / 1M | Kimi positions K2.7 Code as its current coding model; independent normalized benchmarks are not imported. | OpenAI-compatible API; thinking mode required in the documented K2.7 Code quickstart. |
| HighSpeed model ID exists at a higher token price; latency not independently measured here. | Current Kimi coding API lane when Kimi routing fits and 256K context is enough. | Kimi K2.7 Code quickstart [archive], Kimi K2.7 Code pricing [archive], Kimi Code K2.7 release notes [archive], SWE-bench, Berkeley Function Calling Leaderboard | 2026-06-28 |
Status and source type matter as much as a score. GPT-5.6 results are vendor-reported preview evidence; site-owned cost-per-successful-task and repository quality remain not verified.
OpenAI reports that Sol sets a new state of the art on Terminal-Bench 2.1, improves on GPT-5.5 on GeneBench v1, and is competitive with Mythos Preview on ExploitBench while using fewer output tokens. Those results are useful for deciding what to test. OpenAI has not yet published the expanded evaluation suite planned for broader release, and independent reproducible results do not yet support an AIHackers ranking.
Do not combine these benchmark families into one percentage:
- Terminal-Bench 2.1 tests command-line agent tasks.
- GeneBench v1 covers long-horizon genomics and quantitative biology.
- ExploitBench and ExploitGym cover security workflows.
- Artificial Analysis provides a separate independent aggregate for models it has evaluated.
- Cost per successful task still requires the same real task, harness, completion rules, retries, and review process.
Deployment Gate
If your organization receives access:
- Confirm the exact model ID and approved API organization or Codex workspace.
- Record the applicable agreement, retention terms, region, and safeguard behavior.
- Use disposable infrastructure and read-only credentials for the first agent tests.
- Require confirmation for destructive operations, credential movement, uploads, and scope expansion.
- Verify completion from diffs, tests, logs, and external state instead of the model’s summary.
- Measure input, output, cache writes, retries, latency, accepted patches, and human review time.
The preview system card documents low-frequency but serious simulated failures involving destructive scope expansion, credential movement, and unverified completion claims. This evidence supports stronger controls; it does not prove those failures are routine.
What Changes the Recommendation
Move GPT-5.6 from preview to active only after OpenAI publishes broader availability and the target account can call it without a special preview gate. Re-run the comparison when OpenAI publishes stable context/rate-limit specifications, updated safety evidence, or an expanded evaluation suite, or when independent evaluators publish reproducible results.
Sources
- OpenAI: Previewing GPT-5.6 Sol (Archive)
- OpenAI Help Center: A preview of GPT-5.6 Sol, Terra, and Luna (Archive)
- OpenAI: GPT-5.6 Preview System Card (Archive)
- OpenAI: Public model catalog (Archive)
Related links
- /posts/openai-gpt-5-6-sol-limited-preview/
- /posts/frontier-model-access-gates-fable-gpt-5-6/
- /compare/models/premium/
- /models/claude-opus-4-8/
- /models/glm-5.2/
- /models/kimi-k2.7-code/
Last verified: June 28, 2026. Availability, account eligibility, safeguards, pricing, and model specifications can change independently.